1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
28 #include <sys/capability.h>
29 #include <sys/eventfd.h>
31 #include <sys/personality.h>
32 #include <sys/prctl.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
42 #include <security/pam_appl.h>
46 #include <selinux/selinux.h>
54 #include <sys/apparmor.h>
57 #include "sd-messages.h"
60 #include "alloc-util.h"
62 #include "apparmor-util.h"
67 #include "capability-util.h"
68 #include "chown-recursive.h"
69 #include "cpu-set-util.h"
72 #include "errno-list.h"
74 #include "exit-status.h"
77 #include "format-util.h"
79 #include "glob-util.h"
88 #include "namespace.h"
89 #include "parse-util.h"
90 #include "path-util.h"
91 #include "process-util.h"
92 #include "rlimit-util.h"
95 #include "seccomp-util.h"
97 #include "securebits.h"
98 #include "securebits-util.h"
99 #include "selinux-util.h"
100 #include "signal-util.h"
101 #include "smack-util.h"
103 #include "stat-util.h"
104 #include "string-table.h"
105 #include "string-util.h"
107 #include "syslog-util.h"
108 #include "terminal-util.h"
110 #include "user-util.h"
112 #include "utmp-wtmp.h"
114 #define IDLE_TIMEOUT_USEC (5*USEC_PER_SEC)
115 #define IDLE_TIMEOUT2_USEC (1*USEC_PER_SEC)
117 /* This assumes there is a 'tty' group */
118 #define TTY_MODE 0620
120 #define SNDBUF_SIZE (8*1024*1024)
122 static int shift_fds(int fds
[], unsigned n_fds
) {
123 int start
, restart_from
;
128 /* Modifies the fds array! (sorts it) */
138 for (i
= start
; i
< (int) n_fds
; i
++) {
141 /* Already at right index? */
145 nfd
= fcntl(fds
[i
], F_DUPFD
, i
+ 3);
152 /* Hmm, the fd we wanted isn't free? Then
153 * let's remember that and try again from here */
154 if (nfd
!= i
+3 && restart_from
< 0)
158 if (restart_from
< 0)
161 start
= restart_from
;
167 static int flags_fds(const int fds
[], unsigned n_storage_fds
, unsigned n_socket_fds
, bool nonblock
) {
171 n_fds
= n_storage_fds
+ n_socket_fds
;
177 /* Drops/Sets O_NONBLOCK and FD_CLOEXEC from the file flags.
178 * O_NONBLOCK only applies to socket activation though. */
180 for (i
= 0; i
< n_fds
; i
++) {
182 if (i
< n_socket_fds
) {
183 r
= fd_nonblock(fds
[i
], nonblock
);
188 /* We unconditionally drop FD_CLOEXEC from the fds,
189 * since after all we want to pass these fds to our
192 r
= fd_cloexec(fds
[i
], false);
200 static const char *exec_context_tty_path(const ExecContext
*context
) {
203 if (context
->stdio_as_fds
)
206 if (context
->tty_path
)
207 return context
->tty_path
;
209 return "/dev/console";
212 static void exec_context_tty_reset(const ExecContext
*context
, const ExecParameters
*p
) {
217 path
= exec_context_tty_path(context
);
219 if (context
->tty_vhangup
) {
220 if (p
&& p
->stdin_fd
>= 0)
221 (void) terminal_vhangup_fd(p
->stdin_fd
);
223 (void) terminal_vhangup(path
);
226 if (context
->tty_reset
) {
227 if (p
&& p
->stdin_fd
>= 0)
228 (void) reset_terminal_fd(p
->stdin_fd
, true);
230 (void) reset_terminal(path
);
233 if (context
->tty_vt_disallocate
&& path
)
234 (void) vt_disallocate(path
);
237 static bool is_terminal_input(ExecInput i
) {
240 EXEC_INPUT_TTY_FORCE
,
241 EXEC_INPUT_TTY_FAIL
);
244 static bool is_terminal_output(ExecOutput o
) {
247 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
248 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
249 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
);
252 static bool is_syslog_output(ExecOutput o
) {
255 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
);
258 static bool is_kmsg_output(ExecOutput o
) {
261 EXEC_OUTPUT_KMSG_AND_CONSOLE
);
264 static bool exec_context_needs_term(const ExecContext
*c
) {
267 /* Return true if the execution context suggests we should set $TERM to something useful. */
269 if (is_terminal_input(c
->std_input
))
272 if (is_terminal_output(c
->std_output
))
275 if (is_terminal_output(c
->std_error
))
278 return !!c
->tty_path
;
281 static int open_null_as(int flags
, int nfd
) {
286 fd
= open("/dev/null", flags
|O_NOCTTY
);
290 return move_fd(fd
, nfd
, false);
293 static int connect_journal_socket(int fd
, uid_t uid
, gid_t gid
) {
294 static const union sockaddr_union sa
= {
295 .un
.sun_family
= AF_UNIX
,
296 .un
.sun_path
= "/run/systemd/journal/stdout",
298 uid_t olduid
= UID_INVALID
;
299 gid_t oldgid
= GID_INVALID
;
302 if (gid_is_valid(gid
)) {
305 if (setegid(gid
) < 0)
309 if (uid_is_valid(uid
)) {
312 if (seteuid(uid
) < 0) {
318 r
= connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0 ? -errno
: 0;
320 /* If we fail to restore the uid or gid, things will likely
321 fail later on. This should only happen if an LSM interferes. */
323 if (uid_is_valid(uid
))
324 (void) seteuid(olduid
);
327 if (gid_is_valid(gid
))
328 (void) setegid(oldgid
);
333 static int connect_logger_as(
335 const ExecContext
*context
,
336 const ExecParameters
*params
,
347 assert(output
< _EXEC_OUTPUT_MAX
);
351 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
355 r
= connect_journal_socket(fd
, uid
, gid
);
359 if (shutdown(fd
, SHUT_RD
) < 0) {
364 (void) fd_inc_sndbuf(fd
, SNDBUF_SIZE
);
374 context
->syslog_identifier
?: ident
,
375 params
->flags
& EXEC_PASS_LOG_UNIT
? unit
->id
: "",
376 context
->syslog_priority
,
377 !!context
->syslog_level_prefix
,
378 is_syslog_output(output
),
379 is_kmsg_output(output
),
380 is_terminal_output(output
));
382 return move_fd(fd
, nfd
, false);
384 static int open_terminal_as(const char *path
, int flags
, int nfd
) {
390 fd
= open_terminal(path
, flags
| O_NOCTTY
);
394 return move_fd(fd
, nfd
, false);
397 static int acquire_path(const char *path
, int flags
, mode_t mode
) {
398 union sockaddr_union sa
= {
399 .sa
.sa_family
= AF_UNIX
,
405 if (IN_SET(flags
& O_ACCMODE
, O_WRONLY
, O_RDWR
))
408 fd
= open(path
, flags
|O_NOCTTY
, mode
);
412 if (errno
!= ENXIO
) /* ENXIO is returned when we try to open() an AF_UNIX file system socket on Linux */
414 if (strlen(path
) > sizeof(sa
.un
.sun_path
)) /* Too long, can't be a UNIX socket */
417 /* So, it appears the specified path could be an AF_UNIX socket. Let's see if we can connect to it. */
419 fd
= socket(AF_UNIX
, SOCK_STREAM
, 0);
423 strncpy(sa
.un
.sun_path
, path
, sizeof(sa
.un
.sun_path
));
424 if (connect(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0) {
426 return errno
== EINVAL
? -ENXIO
: -errno
; /* Propagate initial error if we get EINVAL, i.e. we have
427 * indication that his wasn't an AF_UNIX socket after all */
430 if ((flags
& O_ACCMODE
) == O_RDONLY
)
431 r
= shutdown(fd
, SHUT_WR
);
432 else if ((flags
& O_ACCMODE
) == O_WRONLY
)
433 r
= shutdown(fd
, SHUT_RD
);
444 static int fixup_input(
445 const ExecContext
*context
,
447 bool apply_tty_stdin
) {
453 std_input
= context
->std_input
;
455 if (is_terminal_input(std_input
) && !apply_tty_stdin
)
456 return EXEC_INPUT_NULL
;
458 if (std_input
== EXEC_INPUT_SOCKET
&& socket_fd
< 0)
459 return EXEC_INPUT_NULL
;
461 if (std_input
== EXEC_INPUT_DATA
&& context
->stdin_data_size
== 0)
462 return EXEC_INPUT_NULL
;
467 static int fixup_output(ExecOutput std_output
, int socket_fd
) {
469 if (std_output
== EXEC_OUTPUT_SOCKET
&& socket_fd
< 0)
470 return EXEC_OUTPUT_INHERIT
;
475 static int setup_input(
476 const ExecContext
*context
,
477 const ExecParameters
*params
,
479 int named_iofds
[3]) {
486 if (params
->stdin_fd
>= 0) {
487 if (dup2(params
->stdin_fd
, STDIN_FILENO
) < 0)
490 /* Try to make this the controlling tty, if it is a tty, and reset it */
491 if (isatty(STDIN_FILENO
)) {
492 (void) ioctl(STDIN_FILENO
, TIOCSCTTY
, context
->std_input
== EXEC_INPUT_TTY_FORCE
);
493 (void) reset_terminal_fd(STDIN_FILENO
, true);
499 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
503 case EXEC_INPUT_NULL
:
504 return open_null_as(O_RDONLY
, STDIN_FILENO
);
507 case EXEC_INPUT_TTY_FORCE
:
508 case EXEC_INPUT_TTY_FAIL
: {
511 fd
= acquire_terminal(exec_context_tty_path(context
),
512 i
== EXEC_INPUT_TTY_FAIL
? ACQUIRE_TERMINAL_TRY
:
513 i
== EXEC_INPUT_TTY_FORCE
? ACQUIRE_TERMINAL_FORCE
:
514 ACQUIRE_TERMINAL_WAIT
,
519 return move_fd(fd
, STDIN_FILENO
, false);
522 case EXEC_INPUT_SOCKET
:
523 assert(socket_fd
>= 0);
525 return dup2(socket_fd
, STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
527 case EXEC_INPUT_NAMED_FD
:
528 assert(named_iofds
[STDIN_FILENO
] >= 0);
530 (void) fd_nonblock(named_iofds
[STDIN_FILENO
], false);
531 return dup2(named_iofds
[STDIN_FILENO
], STDIN_FILENO
) < 0 ? -errno
: STDIN_FILENO
;
533 case EXEC_INPUT_DATA
: {
536 fd
= acquire_data_fd(context
->stdin_data
, context
->stdin_data_size
, 0);
540 return move_fd(fd
, STDIN_FILENO
, false);
543 case EXEC_INPUT_FILE
: {
547 assert(context
->stdio_file
[STDIN_FILENO
]);
549 rw
= (context
->std_output
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDOUT_FILENO
])) ||
550 (context
->std_error
== EXEC_OUTPUT_FILE
&& streq_ptr(context
->stdio_file
[STDIN_FILENO
], context
->stdio_file
[STDERR_FILENO
]));
552 fd
= acquire_path(context
->stdio_file
[STDIN_FILENO
], rw
? O_RDWR
: O_RDONLY
, 0666 & ~context
->umask
);
556 return move_fd(fd
, STDIN_FILENO
, false);
560 assert_not_reached("Unknown input type");
564 static int setup_output(
566 const ExecContext
*context
,
567 const ExecParameters
*params
,
574 dev_t
*journal_stream_dev
,
575 ino_t
*journal_stream_ino
) {
585 assert(journal_stream_dev
);
586 assert(journal_stream_ino
);
588 if (fileno
== STDOUT_FILENO
&& params
->stdout_fd
>= 0) {
590 if (dup2(params
->stdout_fd
, STDOUT_FILENO
) < 0)
593 return STDOUT_FILENO
;
596 if (fileno
== STDERR_FILENO
&& params
->stderr_fd
>= 0) {
597 if (dup2(params
->stderr_fd
, STDERR_FILENO
) < 0)
600 return STDERR_FILENO
;
603 i
= fixup_input(context
, socket_fd
, params
->flags
& EXEC_APPLY_TTY_STDIN
);
604 o
= fixup_output(context
->std_output
, socket_fd
);
606 if (fileno
== STDERR_FILENO
) {
608 e
= fixup_output(context
->std_error
, socket_fd
);
610 /* This expects the input and output are already set up */
612 /* Don't change the stderr file descriptor if we inherit all
613 * the way and are not on a tty */
614 if (e
== EXEC_OUTPUT_INHERIT
&&
615 o
== EXEC_OUTPUT_INHERIT
&&
616 i
== EXEC_INPUT_NULL
&&
617 !is_terminal_input(context
->std_input
) &&
621 /* Duplicate from stdout if possible */
622 if ((e
== o
&& e
!= EXEC_OUTPUT_NAMED_FD
) || e
== EXEC_OUTPUT_INHERIT
)
623 return dup2(STDOUT_FILENO
, fileno
) < 0 ? -errno
: fileno
;
627 } else if (o
== EXEC_OUTPUT_INHERIT
) {
628 /* If input got downgraded, inherit the original value */
629 if (i
== EXEC_INPUT_NULL
&& is_terminal_input(context
->std_input
))
630 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
632 /* If the input is connected to anything that's not a /dev/null or a data fd, inherit that... */
633 if (!IN_SET(i
, EXEC_INPUT_NULL
, EXEC_INPUT_DATA
))
634 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
636 /* If we are not started from PID 1 we just inherit STDOUT from our parent process. */
640 /* We need to open /dev/null here anew, to get the right access mode. */
641 return open_null_as(O_WRONLY
, fileno
);
646 case EXEC_OUTPUT_NULL
:
647 return open_null_as(O_WRONLY
, fileno
);
649 case EXEC_OUTPUT_TTY
:
650 if (is_terminal_input(i
))
651 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
653 /* We don't reset the terminal if this is just about output */
654 return open_terminal_as(exec_context_tty_path(context
), O_WRONLY
, fileno
);
656 case EXEC_OUTPUT_SYSLOG
:
657 case EXEC_OUTPUT_SYSLOG_AND_CONSOLE
:
658 case EXEC_OUTPUT_KMSG
:
659 case EXEC_OUTPUT_KMSG_AND_CONSOLE
:
660 case EXEC_OUTPUT_JOURNAL
:
661 case EXEC_OUTPUT_JOURNAL_AND_CONSOLE
:
662 r
= connect_logger_as(unit
, context
, params
, o
, ident
, fileno
, uid
, gid
);
664 log_unit_warning_errno(unit
, r
, "Failed to connect %s to the journal socket, ignoring: %m", fileno
== STDOUT_FILENO
? "stdout" : "stderr");
665 r
= open_null_as(O_WRONLY
, fileno
);
669 /* If we connected this fd to the journal via a stream, patch the device/inode into the passed
670 * parameters, but only then. This is useful so that we can set $JOURNAL_STREAM that permits
671 * services to detect whether they are connected to the journal or not.
673 * If both stdout and stderr are connected to a stream then let's make sure to store the data
674 * about STDERR as that's usually the best way to do logging. */
676 if (fstat(fileno
, &st
) >= 0 &&
677 (*journal_stream_ino
== 0 || fileno
== STDERR_FILENO
)) {
678 *journal_stream_dev
= st
.st_dev
;
679 *journal_stream_ino
= st
.st_ino
;
684 case EXEC_OUTPUT_SOCKET
:
685 assert(socket_fd
>= 0);
687 return dup2(socket_fd
, fileno
) < 0 ? -errno
: fileno
;
689 case EXEC_OUTPUT_NAMED_FD
:
690 assert(named_iofds
[fileno
] >= 0);
692 (void) fd_nonblock(named_iofds
[fileno
], false);
693 return dup2(named_iofds
[fileno
], fileno
) < 0 ? -errno
: fileno
;
695 case EXEC_OUTPUT_FILE
: {
699 assert(context
->stdio_file
[fileno
]);
701 rw
= context
->std_input
== EXEC_INPUT_FILE
&&
702 streq_ptr(context
->stdio_file
[fileno
], context
->stdio_file
[STDIN_FILENO
]);
705 return dup2(STDIN_FILENO
, fileno
) < 0 ? -errno
: fileno
;
707 fd
= acquire_path(context
->stdio_file
[fileno
], O_WRONLY
, 0666 & ~context
->umask
);
711 return move_fd(fd
, fileno
, false);
715 assert_not_reached("Unknown error type");
719 static int chown_terminal(int fd
, uid_t uid
) {
724 /* Before we chown/chmod the TTY, let's ensure this is actually a tty */
728 /* This might fail. What matters are the results. */
729 (void) fchown(fd
, uid
, -1);
730 (void) fchmod(fd
, TTY_MODE
);
732 if (fstat(fd
, &st
) < 0)
735 if (st
.st_uid
!= uid
|| (st
.st_mode
& 0777) != TTY_MODE
)
741 static int setup_confirm_stdio(const char *vc
, int *_saved_stdin
, int *_saved_stdout
) {
742 _cleanup_close_
int fd
= -1, saved_stdin
= -1, saved_stdout
= -1;
745 assert(_saved_stdin
);
746 assert(_saved_stdout
);
748 saved_stdin
= fcntl(STDIN_FILENO
, F_DUPFD
, 3);
752 saved_stdout
= fcntl(STDOUT_FILENO
, F_DUPFD
, 3);
753 if (saved_stdout
< 0)
756 fd
= acquire_terminal(vc
, ACQUIRE_TERMINAL_WAIT
, DEFAULT_CONFIRM_USEC
);
760 r
= chown_terminal(fd
, getuid());
764 r
= reset_terminal_fd(fd
, true);
768 r
= rearrange_stdio(fd
, fd
, STDERR_FILENO
);
773 *_saved_stdin
= saved_stdin
;
774 *_saved_stdout
= saved_stdout
;
776 saved_stdin
= saved_stdout
= -1;
781 static void write_confirm_error_fd(int err
, int fd
, const Unit
*u
) {
784 if (err
== -ETIMEDOUT
)
785 dprintf(fd
, "Confirmation question timed out for %s, assuming positive response.\n", u
->id
);
788 dprintf(fd
, "Couldn't ask confirmation for %s: %m, assuming positive response.\n", u
->id
);
792 static void write_confirm_error(int err
, const char *vc
, const Unit
*u
) {
793 _cleanup_close_
int fd
= -1;
797 fd
= open_terminal(vc
, O_WRONLY
|O_NOCTTY
|O_CLOEXEC
);
801 write_confirm_error_fd(err
, fd
, u
);
804 static int restore_confirm_stdio(int *saved_stdin
, int *saved_stdout
) {
808 assert(saved_stdout
);
812 if (*saved_stdin
>= 0)
813 if (dup2(*saved_stdin
, STDIN_FILENO
) < 0)
816 if (*saved_stdout
>= 0)
817 if (dup2(*saved_stdout
, STDOUT_FILENO
) < 0)
820 *saved_stdin
= safe_close(*saved_stdin
);
821 *saved_stdout
= safe_close(*saved_stdout
);
827 CONFIRM_PRETEND_FAILURE
= -1,
828 CONFIRM_PRETEND_SUCCESS
= 0,
832 static int ask_for_confirmation(const char *vc
, Unit
*u
, const char *cmdline
) {
833 int saved_stdout
= -1, saved_stdin
= -1, r
;
834 _cleanup_free_
char *e
= NULL
;
837 /* For any internal errors, assume a positive response. */
838 r
= setup_confirm_stdio(vc
, &saved_stdin
, &saved_stdout
);
840 write_confirm_error(r
, vc
, u
);
841 return CONFIRM_EXECUTE
;
844 /* confirm_spawn might have been disabled while we were sleeping. */
845 if (manager_is_confirm_spawn_disabled(u
->manager
)) {
850 e
= ellipsize(cmdline
, 60, 100);
858 r
= ask_char(&c
, "yfshiDjcn", "Execute %s? [y, f, s – h for help] ", e
);
860 write_confirm_error_fd(r
, STDOUT_FILENO
, u
);
867 printf("Resuming normal execution.\n");
868 manager_disable_confirm_spawn();
872 unit_dump(u
, stdout
, " ");
873 continue; /* ask again */
875 printf("Failing execution.\n");
876 r
= CONFIRM_PRETEND_FAILURE
;
879 printf(" c - continue, proceed without asking anymore\n"
880 " D - dump, show the state of the unit\n"
881 " f - fail, don't execute the command and pretend it failed\n"
883 " i - info, show a short summary of the unit\n"
884 " j - jobs, show jobs that are in progress\n"
885 " s - skip, don't execute the command and pretend it succeeded\n"
886 " y - yes, execute the command\n");
887 continue; /* ask again */
889 printf(" Description: %s\n"
892 u
->id
, u
->description
, cmdline
);
893 continue; /* ask again */
895 manager_dump_jobs(u
->manager
, stdout
, " ");
896 continue; /* ask again */
898 /* 'n' was removed in favor of 'f'. */
899 printf("Didn't understand 'n', did you mean 'f'?\n");
900 continue; /* ask again */
902 printf("Skipping execution.\n");
903 r
= CONFIRM_PRETEND_SUCCESS
;
909 assert_not_reached("Unhandled choice");
915 restore_confirm_stdio(&saved_stdin
, &saved_stdout
);
919 static int get_fixed_user(const ExecContext
*c
, const char **user
,
920 uid_t
*uid
, gid_t
*gid
,
921 const char **home
, const char **shell
) {
930 /* Note that we don't set $HOME or $SHELL if they are not particularly enlightening anyway
931 * (i.e. are "/" or "/bin/nologin"). */
934 r
= get_user_creds_clean(&name
, uid
, gid
, home
, shell
);
942 static int get_fixed_group(const ExecContext
*c
, const char **group
, gid_t
*gid
) {
952 r
= get_group_creds(&name
, gid
);
960 static int get_supplementary_groups(const ExecContext
*c
, const char *user
,
961 const char *group
, gid_t gid
,
962 gid_t
**supplementary_gids
, int *ngids
) {
966 bool keep_groups
= false;
967 gid_t
*groups
= NULL
;
968 _cleanup_free_ gid_t
*l_gids
= NULL
;
973 * If user is given, then lookup GID and supplementary groups list.
974 * We avoid NSS lookups for gid=0. Also we have to initialize groups
975 * here and as early as possible so we keep the list of supplementary
976 * groups of the caller.
978 if (user
&& gid_is_valid(gid
) && gid
!= 0) {
979 /* First step, initialize groups from /etc/groups */
980 if (initgroups(user
, gid
) < 0)
986 if (strv_isempty(c
->supplementary_groups
))
990 * If SupplementaryGroups= was passed then NGROUPS_MAX has to
991 * be positive, otherwise fail.
994 ngroups_max
= (int) sysconf(_SC_NGROUPS_MAX
);
995 if (ngroups_max
<= 0) {
999 return -EOPNOTSUPP
; /* For all other values */
1002 l_gids
= new(gid_t
, ngroups_max
);
1008 * Lookup the list of groups that the user belongs to, we
1009 * avoid NSS lookups here too for gid=0.
1012 if (getgrouplist(user
, gid
, l_gids
, &k
) < 0)
1017 STRV_FOREACH(i
, c
->supplementary_groups
) {
1020 if (k
>= ngroups_max
)
1024 r
= get_group_creds(&g
, l_gids
+k
);
1032 * Sets ngids to zero to drop all supplementary groups, happens
1033 * when we are under root and SupplementaryGroups= is empty.
1040 /* Otherwise get the final list of supplementary groups */
1041 groups
= memdup(l_gids
, sizeof(gid_t
) * k
);
1045 *supplementary_gids
= groups
;
1053 static int enforce_groups(gid_t gid
, const gid_t
*supplementary_gids
, int ngids
) {
1056 /* Handle SupplementaryGroups= if it is not empty */
1058 r
= maybe_setgroups(ngids
, supplementary_gids
);
1063 if (gid_is_valid(gid
)) {
1064 /* Then set our gids */
1065 if (setresgid(gid
, gid
, gid
) < 0)
1072 static int enforce_user(const ExecContext
*context
, uid_t uid
) {
1075 if (!uid_is_valid(uid
))
1078 /* Sets (but doesn't look up) the uid and make sure we keep the
1079 * capabilities while doing so. */
1081 if (context
->capability_ambient_set
!= 0) {
1083 /* First step: If we need to keep capabilities but
1084 * drop privileges we need to make sure we keep our
1085 * caps, while we drop privileges. */
1087 int sb
= context
->secure_bits
| 1<<SECURE_KEEP_CAPS
;
1089 if (prctl(PR_GET_SECUREBITS
) != sb
)
1090 if (prctl(PR_SET_SECUREBITS
, sb
) < 0)
1095 /* Second step: actually set the uids */
1096 if (setresuid(uid
, uid
, uid
) < 0)
1099 /* At this point we should have all necessary capabilities but
1100 are otherwise a normal user. However, the caps might got
1101 corrupted due to the setresuid() so we need clean them up
1102 later. This is done outside of this call. */
1109 static int null_conv(
1111 const struct pam_message
**msg
,
1112 struct pam_response
**resp
,
1113 void *appdata_ptr
) {
1115 /* We don't support conversations */
1117 return PAM_CONV_ERR
;
1122 static int setup_pam(
1129 int fds
[], unsigned n_fds
) {
1133 static const struct pam_conv conv
= {
1138 _cleanup_(barrier_destroy
) Barrier barrier
= BARRIER_NULL
;
1139 pam_handle_t
*handle
= NULL
;
1141 int pam_code
= PAM_SUCCESS
, r
;
1142 char **nv
, **e
= NULL
;
1143 bool close_session
= false;
1144 pid_t pam_pid
= 0, parent_pid
;
1151 /* We set up PAM in the parent process, then fork. The child
1152 * will then stay around until killed via PR_GET_PDEATHSIG or
1153 * systemd via the cgroup logic. It will then remove the PAM
1154 * session again. The parent process will exec() the actual
1155 * daemon. We do things this way to ensure that the main PID
1156 * of the daemon is the one we initially fork()ed. */
1158 r
= barrier_create(&barrier
);
1162 if (log_get_max_level() < LOG_DEBUG
)
1163 flags
|= PAM_SILENT
;
1165 pam_code
= pam_start(name
, user
, &conv
, &handle
);
1166 if (pam_code
!= PAM_SUCCESS
) {
1172 pam_code
= pam_set_item(handle
, PAM_TTY
, tty
);
1173 if (pam_code
!= PAM_SUCCESS
)
1177 STRV_FOREACH(nv
, *env
) {
1178 pam_code
= pam_putenv(handle
, *nv
);
1179 if (pam_code
!= PAM_SUCCESS
)
1183 pam_code
= pam_acct_mgmt(handle
, flags
);
1184 if (pam_code
!= PAM_SUCCESS
)
1187 pam_code
= pam_open_session(handle
, flags
);
1188 if (pam_code
!= PAM_SUCCESS
)
1191 close_session
= true;
1193 e
= pam_getenvlist(handle
);
1195 pam_code
= PAM_BUF_ERR
;
1199 /* Block SIGTERM, so that we know that it won't get lost in
1202 assert_se(sigprocmask_many(SIG_BLOCK
, &old_ss
, SIGTERM
, -1) >= 0);
1204 parent_pid
= getpid_cached();
1206 r
= safe_fork("(sd-pam)", 0, &pam_pid
);
1210 int sig
, ret
= EXIT_PAM
;
1212 /* The child's job is to reset the PAM session on
1214 barrier_set_role(&barrier
, BARRIER_CHILD
);
1216 /* Make sure we don't keep open the passed fds in this child. We assume that otherwise only those fds
1217 * are open here that have been opened by PAM. */
1218 (void) close_many(fds
, n_fds
);
1220 /* Drop privileges - we don't need any to pam_close_session
1221 * and this will make PR_SET_PDEATHSIG work in most cases.
1222 * If this fails, ignore the error - but expect sd-pam threads
1223 * to fail to exit normally */
1225 r
= maybe_setgroups(0, NULL
);
1227 log_warning_errno(r
, "Failed to setgroups() in sd-pam: %m");
1228 if (setresgid(gid
, gid
, gid
) < 0)
1229 log_warning_errno(errno
, "Failed to setresgid() in sd-pam: %m");
1230 if (setresuid(uid
, uid
, uid
) < 0)
1231 log_warning_errno(errno
, "Failed to setresuid() in sd-pam: %m");
1233 (void) ignore_signals(SIGPIPE
, -1);
1235 /* Wait until our parent died. This will only work if
1236 * the above setresuid() succeeds, otherwise the kernel
1237 * will not allow unprivileged parents kill their privileged
1238 * children this way. We rely on the control groups kill logic
1239 * to do the rest for us. */
1240 if (prctl(PR_SET_PDEATHSIG
, SIGTERM
) < 0)
1243 /* Tell the parent that our setup is done. This is especially
1244 * important regarding dropping privileges. Otherwise, unit
1245 * setup might race against our setresuid(2) call.
1247 * If the parent aborted, we'll detect this below, hence ignore
1248 * return failure here. */
1249 (void) barrier_place(&barrier
);
1251 /* Check if our parent process might already have died? */
1252 if (getppid() == parent_pid
) {
1255 assert_se(sigemptyset(&ss
) >= 0);
1256 assert_se(sigaddset(&ss
, SIGTERM
) >= 0);
1259 if (sigwait(&ss
, &sig
) < 0) {
1266 assert(sig
== SIGTERM
);
1271 /* If our parent died we'll end the session */
1272 if (getppid() != parent_pid
) {
1273 pam_code
= pam_close_session(handle
, flags
);
1274 if (pam_code
!= PAM_SUCCESS
)
1281 pam_end(handle
, pam_code
| flags
);
1285 barrier_set_role(&barrier
, BARRIER_PARENT
);
1287 /* If the child was forked off successfully it will do all the
1288 * cleanups, so forget about the handle here. */
1291 /* Unblock SIGTERM again in the parent */
1292 assert_se(sigprocmask(SIG_SETMASK
, &old_ss
, NULL
) >= 0);
1294 /* We close the log explicitly here, since the PAM modules
1295 * might have opened it, but we don't want this fd around. */
1298 /* Synchronously wait for the child to initialize. We don't care for
1299 * errors as we cannot recover. However, warn loudly if it happens. */
1300 if (!barrier_place_and_sync(&barrier
))
1301 log_error("PAM initialization failed");
1309 if (pam_code
!= PAM_SUCCESS
) {
1310 log_error("PAM failed: %s", pam_strerror(handle
, pam_code
));
1311 r
= -EPERM
; /* PAM errors do not map to errno */
1313 log_error_errno(r
, "PAM failed: %m");
1317 pam_code
= pam_close_session(handle
, flags
);
1319 pam_end(handle
, pam_code
| flags
);
1331 static void rename_process_from_path(const char *path
) {
1332 char process_name
[11];
1336 /* This resulting string must fit in 10 chars (i.e. the length
1337 * of "/sbin/init") to look pretty in /bin/ps */
1341 rename_process("(...)");
1347 /* The end of the process name is usually more
1348 * interesting, since the first bit might just be
1354 process_name
[0] = '(';
1355 memcpy(process_name
+1, p
, l
);
1356 process_name
[1+l
] = ')';
1357 process_name
[1+l
+1] = 0;
1359 rename_process(process_name
);
1362 static bool context_has_address_families(const ExecContext
*c
) {
1365 return c
->address_families_whitelist
||
1366 !set_isempty(c
->address_families
);
1369 static bool context_has_syscall_filters(const ExecContext
*c
) {
1372 return c
->syscall_whitelist
||
1373 !hashmap_isempty(c
->syscall_filter
);
1376 static bool context_has_no_new_privileges(const ExecContext
*c
) {
1379 if (c
->no_new_privileges
)
1382 if (have_effective_cap(CAP_SYS_ADMIN
)) /* if we are privileged, we don't need NNP */
1385 /* We need NNP if we have any form of seccomp and are unprivileged */
1386 return context_has_address_families(c
) ||
1387 c
->memory_deny_write_execute
||
1388 c
->restrict_realtime
||
1389 exec_context_restrict_namespaces_set(c
) ||
1390 c
->protect_kernel_tunables
||
1391 c
->protect_kernel_modules
||
1392 c
->private_devices
||
1393 context_has_syscall_filters(c
) ||
1394 !set_isempty(c
->syscall_archs
) ||
1395 c
->lock_personality
;
1400 static bool skip_seccomp_unavailable(const Unit
* u
, const char* msg
) {
1402 if (is_seccomp_available())
1405 log_unit_debug(u
, "SECCOMP features not detected in the kernel, skipping %s", msg
);
1409 static int apply_syscall_filter(const Unit
* u
, const ExecContext
*c
, bool needs_ambient_hack
) {
1410 uint32_t negative_action
, default_action
, action
;
1416 if (!context_has_syscall_filters(c
))
1419 if (skip_seccomp_unavailable(u
, "SystemCallFilter="))
1422 negative_action
= c
->syscall_errno
== 0 ? SCMP_ACT_KILL
: SCMP_ACT_ERRNO(c
->syscall_errno
);
1424 if (c
->syscall_whitelist
) {
1425 default_action
= negative_action
;
1426 action
= SCMP_ACT_ALLOW
;
1428 default_action
= SCMP_ACT_ALLOW
;
1429 action
= negative_action
;
1432 if (needs_ambient_hack
) {
1433 r
= seccomp_filter_set_add(c
->syscall_filter
, c
->syscall_whitelist
, syscall_filter_sets
+ SYSCALL_FILTER_SET_SETUID
);
1438 return seccomp_load_syscall_filter_set_raw(default_action
, c
->syscall_filter
, action
);
1441 static int apply_syscall_archs(const Unit
*u
, const ExecContext
*c
) {
1445 if (set_isempty(c
->syscall_archs
))
1448 if (skip_seccomp_unavailable(u
, "SystemCallArchitectures="))
1451 return seccomp_restrict_archs(c
->syscall_archs
);
1454 static int apply_address_families(const Unit
* u
, const ExecContext
*c
) {
1458 if (!context_has_address_families(c
))
1461 if (skip_seccomp_unavailable(u
, "RestrictAddressFamilies="))
1464 return seccomp_restrict_address_families(c
->address_families
, c
->address_families_whitelist
);
1467 static int apply_memory_deny_write_execute(const Unit
* u
, const ExecContext
*c
) {
1471 if (!c
->memory_deny_write_execute
)
1474 if (skip_seccomp_unavailable(u
, "MemoryDenyWriteExecute="))
1477 return seccomp_memory_deny_write_execute();
1480 static int apply_restrict_realtime(const Unit
* u
, const ExecContext
*c
) {
1484 if (!c
->restrict_realtime
)
1487 if (skip_seccomp_unavailable(u
, "RestrictRealtime="))
1490 return seccomp_restrict_realtime();
1493 static int apply_protect_sysctl(const Unit
*u
, const ExecContext
*c
) {
1497 /* Turn off the legacy sysctl() system call. Many distributions turn this off while building the kernel, but
1498 * let's protect even those systems where this is left on in the kernel. */
1500 if (!c
->protect_kernel_tunables
)
1503 if (skip_seccomp_unavailable(u
, "ProtectKernelTunables="))
1506 return seccomp_protect_sysctl();
1509 static int apply_protect_kernel_modules(const Unit
*u
, const ExecContext
*c
) {
1513 /* Turn off module syscalls on ProtectKernelModules=yes */
1515 if (!c
->protect_kernel_modules
)
1518 if (skip_seccomp_unavailable(u
, "ProtectKernelModules="))
1521 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_MODULE
, SCMP_ACT_ERRNO(EPERM
));
1524 static int apply_private_devices(const Unit
*u
, const ExecContext
*c
) {
1528 /* If PrivateDevices= is set, also turn off iopl and all @raw-io syscalls. */
1530 if (!c
->private_devices
)
1533 if (skip_seccomp_unavailable(u
, "PrivateDevices="))
1536 return seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW
, syscall_filter_sets
+ SYSCALL_FILTER_SET_RAW_IO
, SCMP_ACT_ERRNO(EPERM
));
1539 static int apply_restrict_namespaces(const Unit
*u
, const ExecContext
*c
) {
1543 if (!exec_context_restrict_namespaces_set(c
))
1546 if (skip_seccomp_unavailable(u
, "RestrictNamespaces="))
1549 return seccomp_restrict_namespaces(c
->restrict_namespaces
);
1552 static int apply_lock_personality(const Unit
* u
, const ExecContext
*c
) {
1553 unsigned long personality
;
1559 if (!c
->lock_personality
)
1562 if (skip_seccomp_unavailable(u
, "LockPersonality="))
1565 personality
= c
->personality
;
1567 /* If personality is not specified, use either PER_LINUX or PER_LINUX32 depending on what is currently set. */
1568 if (personality
== PERSONALITY_INVALID
) {
1570 r
= opinionated_personality(&personality
);
1575 return seccomp_lock_personality(personality
);
1580 static void do_idle_pipe_dance(int idle_pipe
[4]) {
1583 idle_pipe
[1] = safe_close(idle_pipe
[1]);
1584 idle_pipe
[2] = safe_close(idle_pipe
[2]);
1586 if (idle_pipe
[0] >= 0) {
1589 r
= fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT_USEC
);
1591 if (idle_pipe
[3] >= 0 && r
== 0 /* timeout */) {
1594 /* Signal systemd that we are bored and want to continue. */
1595 n
= write(idle_pipe
[3], "x", 1);
1597 /* Wait for systemd to react to the signal above. */
1598 fd_wait_for_event(idle_pipe
[0], POLLHUP
, IDLE_TIMEOUT2_USEC
);
1601 idle_pipe
[0] = safe_close(idle_pipe
[0]);
1605 idle_pipe
[3] = safe_close(idle_pipe
[3]);
1608 static int build_environment(
1610 const ExecContext
*c
,
1611 const ExecParameters
*p
,
1614 const char *username
,
1616 dev_t journal_stream_dev
,
1617 ino_t journal_stream_ino
,
1620 _cleanup_strv_free_
char **our_env
= NULL
;
1628 our_env
= new0(char*, 14);
1633 _cleanup_free_
char *joined
= NULL
;
1635 if (asprintf(&x
, "LISTEN_PID="PID_FMT
, getpid_cached()) < 0)
1637 our_env
[n_env
++] = x
;
1639 if (asprintf(&x
, "LISTEN_FDS=%u", n_fds
) < 0)
1641 our_env
[n_env
++] = x
;
1643 joined
= strv_join(p
->fd_names
, ":");
1647 x
= strjoin("LISTEN_FDNAMES=", joined
);
1650 our_env
[n_env
++] = x
;
1653 if ((p
->flags
& EXEC_SET_WATCHDOG
) && p
->watchdog_usec
> 0) {
1654 if (asprintf(&x
, "WATCHDOG_PID="PID_FMT
, getpid_cached()) < 0)
1656 our_env
[n_env
++] = x
;
1658 if (asprintf(&x
, "WATCHDOG_USEC="USEC_FMT
, p
->watchdog_usec
) < 0)
1660 our_env
[n_env
++] = x
;
1663 /* If this is D-Bus, tell the nss-systemd module, since it relies on being able to use D-Bus look up dynamic
1664 * users via PID 1, possibly dead-locking the dbus daemon. This way it will not use D-Bus to resolve names, but
1665 * check the database directly. */
1666 if (p
->flags
& EXEC_NSS_BYPASS_BUS
) {
1667 x
= strdup("SYSTEMD_NSS_BYPASS_BUS=1");
1670 our_env
[n_env
++] = x
;
1674 x
= strappend("HOME=", home
);
1677 our_env
[n_env
++] = x
;
1681 x
= strappend("LOGNAME=", username
);
1684 our_env
[n_env
++] = x
;
1686 x
= strappend("USER=", username
);
1689 our_env
[n_env
++] = x
;
1693 x
= strappend("SHELL=", shell
);
1696 our_env
[n_env
++] = x
;
1699 if (!sd_id128_is_null(u
->invocation_id
)) {
1700 if (asprintf(&x
, "INVOCATION_ID=" SD_ID128_FORMAT_STR
, SD_ID128_FORMAT_VAL(u
->invocation_id
)) < 0)
1703 our_env
[n_env
++] = x
;
1706 if (exec_context_needs_term(c
)) {
1707 const char *tty_path
, *term
= NULL
;
1709 tty_path
= exec_context_tty_path(c
);
1711 /* If we are forked off PID 1 and we are supposed to operate on /dev/console, then let's try to inherit
1712 * the $TERM set for PID 1. This is useful for containers so that the $TERM the container manager
1713 * passes to PID 1 ends up all the way in the console login shown. */
1715 if (path_equal(tty_path
, "/dev/console") && getppid() == 1)
1716 term
= getenv("TERM");
1718 term
= default_term_for_tty(tty_path
);
1720 x
= strappend("TERM=", term
);
1723 our_env
[n_env
++] = x
;
1726 if (journal_stream_dev
!= 0 && journal_stream_ino
!= 0) {
1727 if (asprintf(&x
, "JOURNAL_STREAM=" DEV_FMT
":" INO_FMT
, journal_stream_dev
, journal_stream_ino
) < 0)
1730 our_env
[n_env
++] = x
;
1733 our_env
[n_env
++] = NULL
;
1734 assert(n_env
<= 12);
1736 *ret
= TAKE_PTR(our_env
);
1741 static int build_pass_environment(const ExecContext
*c
, char ***ret
) {
1742 _cleanup_strv_free_
char **pass_env
= NULL
;
1743 size_t n_env
= 0, n_bufsize
= 0;
1746 STRV_FOREACH(i
, c
->pass_environment
) {
1747 _cleanup_free_
char *x
= NULL
;
1753 x
= strjoin(*i
, "=", v
);
1757 if (!GREEDY_REALLOC(pass_env
, n_bufsize
, n_env
+ 2))
1760 pass_env
[n_env
++] = TAKE_PTR(x
);
1761 pass_env
[n_env
] = NULL
;
1764 *ret
= TAKE_PTR(pass_env
);
1769 static bool exec_needs_mount_namespace(
1770 const ExecContext
*context
,
1771 const ExecParameters
*params
,
1772 const ExecRuntime
*runtime
) {
1777 if (context
->root_image
)
1780 if (!strv_isempty(context
->read_write_paths
) ||
1781 !strv_isempty(context
->read_only_paths
) ||
1782 !strv_isempty(context
->inaccessible_paths
))
1785 if (context
->n_bind_mounts
> 0)
1788 if (context
->n_temporary_filesystems
> 0)
1791 if (context
->mount_flags
!= 0)
1794 if (context
->private_tmp
&& runtime
&& (runtime
->tmp_dir
|| runtime
->var_tmp_dir
))
1797 if (context
->private_devices
||
1798 context
->protect_system
!= PROTECT_SYSTEM_NO
||
1799 context
->protect_home
!= PROTECT_HOME_NO
||
1800 context
->protect_kernel_tunables
||
1801 context
->protect_kernel_modules
||
1802 context
->protect_control_groups
)
1805 if (context
->mount_apivfs
&& (context
->root_image
|| context
->root_directory
))
1808 if (context
->dynamic_user
&&
1809 (!strv_isempty(context
->directories
[EXEC_DIRECTORY_STATE
].paths
) ||
1810 !strv_isempty(context
->directories
[EXEC_DIRECTORY_CACHE
].paths
) ||
1811 !strv_isempty(context
->directories
[EXEC_DIRECTORY_LOGS
].paths
)))
1817 static int setup_private_users(uid_t uid
, gid_t gid
) {
1818 _cleanup_free_
char *uid_map
= NULL
, *gid_map
= NULL
;
1819 _cleanup_close_pair_
int errno_pipe
[2] = { -1, -1 };
1820 _cleanup_close_
int unshare_ready_fd
= -1;
1821 _cleanup_(sigkill_waitp
) pid_t pid
= 0;
1826 /* Set up a user namespace and map root to root, the selected UID/GID to itself, and everything else to
1827 * nobody. In order to be able to write this mapping we need CAP_SETUID in the original user namespace, which
1828 * we however lack after opening the user namespace. To work around this we fork() a temporary child process,
1829 * which waits for the parent to create the new user namespace while staying in the original namespace. The
1830 * child then writes the UID mapping, under full privileges. The parent waits for the child to finish and
1831 * continues execution normally. */
1833 if (uid
!= 0 && uid_is_valid(uid
)) {
1834 r
= asprintf(&uid_map
,
1835 "0 0 1\n" /* Map root → root */
1836 UID_FMT
" " UID_FMT
" 1\n", /* Map $UID → $UID */
1841 uid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1846 if (gid
!= 0 && gid_is_valid(gid
)) {
1847 r
= asprintf(&gid_map
,
1848 "0 0 1\n" /* Map root → root */
1849 GID_FMT
" " GID_FMT
" 1\n", /* Map $GID → $GID */
1854 gid_map
= strdup("0 0 1\n"); /* The case where the above is the same */
1859 /* Create a communication channel so that the parent can tell the child when it finished creating the user
1861 unshare_ready_fd
= eventfd(0, EFD_CLOEXEC
);
1862 if (unshare_ready_fd
< 0)
1865 /* Create a communication channel so that the child can tell the parent a proper error code in case it
1867 if (pipe2(errno_pipe
, O_CLOEXEC
) < 0)
1870 r
= safe_fork("(sd-userns)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
, &pid
);
1874 _cleanup_close_
int fd
= -1;
1878 /* Child process, running in the original user namespace. Let's update the parent's UID/GID map from
1879 * here, after the parent opened its own user namespace. */
1882 errno_pipe
[0] = safe_close(errno_pipe
[0]);
1884 /* Wait until the parent unshared the user namespace */
1885 if (read(unshare_ready_fd
, &c
, sizeof(c
)) < 0) {
1890 /* Disable the setgroups() system call in the child user namespace, for good. */
1891 a
= procfs_file_alloca(ppid
, "setgroups");
1892 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1894 if (errno
!= ENOENT
) {
1899 /* If the file is missing the kernel is too old, let's continue anyway. */
1901 if (write(fd
, "deny\n", 5) < 0) {
1906 fd
= safe_close(fd
);
1909 /* First write the GID map */
1910 a
= procfs_file_alloca(ppid
, "gid_map");
1911 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1916 if (write(fd
, gid_map
, strlen(gid_map
)) < 0) {
1920 fd
= safe_close(fd
);
1922 /* The write the UID map */
1923 a
= procfs_file_alloca(ppid
, "uid_map");
1924 fd
= open(a
, O_WRONLY
|O_CLOEXEC
);
1929 if (write(fd
, uid_map
, strlen(uid_map
)) < 0) {
1934 _exit(EXIT_SUCCESS
);
1937 (void) write(errno_pipe
[1], &r
, sizeof(r
));
1938 _exit(EXIT_FAILURE
);
1941 errno_pipe
[1] = safe_close(errno_pipe
[1]);
1943 if (unshare(CLONE_NEWUSER
) < 0)
1946 /* Let the child know that the namespace is ready now */
1947 if (write(unshare_ready_fd
, &c
, sizeof(c
)) < 0)
1950 /* Try to read an error code from the child */
1951 n
= read(errno_pipe
[0], &r
, sizeof(r
));
1954 if (n
== sizeof(r
)) { /* an error code was sent to us */
1959 if (n
!= 0) /* on success we should have read 0 bytes */
1962 r
= wait_for_terminate_and_check("(sd-userns)", pid
, 0);
1966 if (r
!= EXIT_SUCCESS
) /* If something strange happened with the child, let's consider this fatal, too */
1972 static int setup_exec_directory(
1973 const ExecContext
*context
,
1974 const ExecParameters
*params
,
1977 ExecDirectoryType type
,
1980 static const int exit_status_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
1981 [EXEC_DIRECTORY_RUNTIME
] = EXIT_RUNTIME_DIRECTORY
,
1982 [EXEC_DIRECTORY_STATE
] = EXIT_STATE_DIRECTORY
,
1983 [EXEC_DIRECTORY_CACHE
] = EXIT_CACHE_DIRECTORY
,
1984 [EXEC_DIRECTORY_LOGS
] = EXIT_LOGS_DIRECTORY
,
1985 [EXEC_DIRECTORY_CONFIGURATION
] = EXIT_CONFIGURATION_DIRECTORY
,
1992 assert(type
>= 0 && type
< _EXEC_DIRECTORY_TYPE_MAX
);
1993 assert(exit_status
);
1995 if (!params
->prefix
[type
])
1998 if (params
->flags
& EXEC_CHOWN_DIRECTORIES
) {
1999 if (!uid_is_valid(uid
))
2001 if (!gid_is_valid(gid
))
2005 STRV_FOREACH(rt
, context
->directories
[type
].paths
) {
2006 _cleanup_free_
char *p
= NULL
, *pp
= NULL
;
2008 p
= strjoin(params
->prefix
[type
], "/", *rt
);
2014 r
= mkdir_parents_label(p
, 0755);
2018 if (context
->dynamic_user
&&
2019 !IN_SET(type
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2020 _cleanup_free_
char *private_root
= NULL
, *relative
= NULL
, *parent
= NULL
;
2022 /* So, here's one extra complication when dealing with DynamicUser=1 units. In that case we
2023 * want to avoid leaving a directory around fully accessible that is owned by a dynamic user
2024 * whose UID is later on reused. To lock this down we use the same trick used by container
2025 * managers to prohibit host users to get access to files of the same UID in containers: we
2026 * place everything inside a directory that has an access mode of 0700 and is owned root:root,
2027 * so that it acts as security boundary for unprivileged host code. We then use fs namespacing
2028 * to make this directory permeable for the service itself.
2030 * Specifically: for a service which wants a special directory "foo/" we first create a
2031 * directory "private/" with access mode 0700 owned by root:root. Then we place "foo" inside of
2032 * that directory (i.e. "private/foo/"), and make "foo" a symlink to "private/foo". This way,
2033 * privileged host users can access "foo/" as usual, but unprivileged host users can't look
2034 * into it. Inside of the namespaceof the container "private/" is replaced by a more liberally
2035 * accessible tmpfs, into which the host's "private/foo/" is mounted under the same name, thus
2036 * disabling the access boundary for the service and making sure it only gets access to the
2037 * dirs it needs but no others. Tricky? Yes, absolutely, but it works!
2039 * Note that we don't do this for EXEC_DIRECTORY_CONFIGURATION as that's assumed not to be
2040 * owned by the service itself.
2041 * Also, note that we don't do this for EXEC_DIRECTORY_RUNTIME as that's often used for sharing
2042 * files or sockets with other services. */
2044 private_root
= strjoin(params
->prefix
[type
], "/private");
2045 if (!private_root
) {
2050 /* First set up private root if it doesn't exist yet, with access mode 0700 and owned by root:root */
2051 r
= mkdir_safe_label(private_root
, 0700, 0, 0, MKDIR_WARN_MODE
);
2055 pp
= strjoin(private_root
, "/", *rt
);
2061 /* Create all directories between the configured directory and this private root, and mark them 0755 */
2062 r
= mkdir_parents_label(pp
, 0755);
2066 if (is_dir(p
, false) > 0 &&
2067 (laccess(pp
, F_OK
) < 0 && errno
== ENOENT
)) {
2069 /* Hmm, the private directory doesn't exist yet, but the normal one exists? If so, move
2070 * it over. Most likely the service has been upgraded from one that didn't use
2071 * DynamicUser=1, to one that does. */
2073 if (rename(p
, pp
) < 0) {
2078 /* Otherwise, create the actual directory for the service */
2080 r
= mkdir_label(pp
, context
->directories
[type
].mode
);
2081 if (r
< 0 && r
!= -EEXIST
)
2085 parent
= dirname_malloc(p
);
2091 r
= path_make_relative(parent
, pp
, &relative
);
2095 /* And link it up from the original place */
2096 r
= symlink_idempotent(relative
, p
);
2100 /* Lock down the access mode */
2101 if (chmod(pp
, context
->directories
[type
].mode
) < 0) {
2106 r
= mkdir_label(p
, context
->directories
[type
].mode
);
2113 /* Don't change the owner of the configuration directory, as in the common case it is not written to by
2114 * a service, and shall not be writable. */
2115 if (type
== EXEC_DIRECTORY_CONFIGURATION
)
2118 /* Then, change the ownership of the whole tree, if necessary */
2119 r
= path_chown_recursive(pp
?: p
, uid
, gid
);
2127 *exit_status
= exit_status_table
[type
];
2132 static int setup_smack(
2133 const ExecContext
*context
,
2134 const ExecCommand
*command
) {
2141 if (context
->smack_process_label
) {
2142 r
= mac_smack_apply_pid(0, context
->smack_process_label
);
2146 #ifdef SMACK_DEFAULT_PROCESS_LABEL
2148 _cleanup_free_
char *exec_label
= NULL
;
2150 r
= mac_smack_read(command
->path
, SMACK_ATTR_EXEC
, &exec_label
);
2151 if (r
< 0 && !IN_SET(r
, -ENODATA
, -EOPNOTSUPP
))
2154 r
= mac_smack_apply_pid(0, exec_label
? : SMACK_DEFAULT_PROCESS_LABEL
);
2164 static int compile_bind_mounts(
2165 const ExecContext
*context
,
2166 const ExecParameters
*params
,
2167 BindMount
**ret_bind_mounts
,
2168 unsigned *ret_n_bind_mounts
,
2169 char ***ret_empty_directories
) {
2171 _cleanup_strv_free_
char **empty_directories
= NULL
;
2172 BindMount
*bind_mounts
;
2173 unsigned n
, h
= 0, i
;
2174 ExecDirectoryType t
;
2179 assert(ret_bind_mounts
);
2180 assert(ret_n_bind_mounts
);
2181 assert(ret_empty_directories
);
2183 n
= context
->n_bind_mounts
;
2184 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2185 if (!params
->prefix
[t
])
2188 n
+= strv_length(context
->directories
[t
].paths
);
2192 *ret_bind_mounts
= NULL
;
2193 *ret_n_bind_mounts
= 0;
2194 *ret_empty_directories
= NULL
;
2198 bind_mounts
= new(BindMount
, n
);
2202 for (i
= 0; i
< context
->n_bind_mounts
; i
++) {
2203 BindMount
*item
= context
->bind_mounts
+ i
;
2206 s
= strdup(item
->source
);
2212 d
= strdup(item
->destination
);
2219 bind_mounts
[h
++] = (BindMount
) {
2222 .read_only
= item
->read_only
,
2223 .recursive
= item
->recursive
,
2224 .ignore_enoent
= item
->ignore_enoent
,
2228 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2231 if (!params
->prefix
[t
])
2234 if (strv_isempty(context
->directories
[t
].paths
))
2237 if (context
->dynamic_user
&&
2238 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
)) {
2241 /* So this is for a dynamic user, and we need to make sure the process can access its own
2242 * directory. For that we overmount the usually inaccessible "private" subdirectory with a
2243 * tmpfs that makes it accessible and is empty except for the submounts we do this for. */
2245 private_root
= strjoin(params
->prefix
[t
], "/private");
2246 if (!private_root
) {
2251 r
= strv_consume(&empty_directories
, private_root
);
2256 STRV_FOREACH(suffix
, context
->directories
[t
].paths
) {
2259 if (context
->dynamic_user
&&
2260 !IN_SET(t
, EXEC_DIRECTORY_RUNTIME
, EXEC_DIRECTORY_CONFIGURATION
))
2261 s
= strjoin(params
->prefix
[t
], "/private/", *suffix
);
2263 s
= strjoin(params
->prefix
[t
], "/", *suffix
);
2276 bind_mounts
[h
++] = (BindMount
) {
2281 .ignore_enoent
= false,
2288 *ret_bind_mounts
= bind_mounts
;
2289 *ret_n_bind_mounts
= n
;
2290 *ret_empty_directories
= TAKE_PTR(empty_directories
);
2295 bind_mount_free_many(bind_mounts
, h
);
2299 static int apply_mount_namespace(
2301 const ExecCommand
*command
,
2302 const ExecContext
*context
,
2303 const ExecParameters
*params
,
2304 const ExecRuntime
*runtime
) {
2306 _cleanup_strv_free_
char **empty_directories
= NULL
;
2307 char *tmp
= NULL
, *var
= NULL
;
2308 const char *root_dir
= NULL
, *root_image
= NULL
;
2309 NamespaceInfo ns_info
= {
2310 .ignore_protect_paths
= false,
2311 .private_dev
= context
->private_devices
,
2312 .protect_control_groups
= context
->protect_control_groups
,
2313 .protect_kernel_tunables
= context
->protect_kernel_tunables
,
2314 .protect_kernel_modules
= context
->protect_kernel_modules
,
2315 .mount_apivfs
= context
->mount_apivfs
,
2317 bool needs_sandboxing
;
2318 BindMount
*bind_mounts
= NULL
;
2319 unsigned n_bind_mounts
= 0;
2324 /* The runtime struct only contains the parent of the private /tmp,
2325 * which is non-accessible to world users. Inside of it there's a /tmp
2326 * that is sticky, and that's the one we want to use here. */
2328 if (context
->private_tmp
&& runtime
) {
2329 if (runtime
->tmp_dir
)
2330 tmp
= strjoina(runtime
->tmp_dir
, "/tmp");
2331 if (runtime
->var_tmp_dir
)
2332 var
= strjoina(runtime
->var_tmp_dir
, "/tmp");
2335 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2336 root_image
= context
->root_image
;
2339 root_dir
= context
->root_directory
;
2342 r
= compile_bind_mounts(context
, params
, &bind_mounts
, &n_bind_mounts
, &empty_directories
);
2347 * If DynamicUser=no and RootDirectory= is set then lets pass a relaxed
2348 * sandbox info, otherwise enforce it, don't ignore protected paths and
2349 * fail if we are enable to apply the sandbox inside the mount namespace.
2351 if (!context
->dynamic_user
&& root_dir
)
2352 ns_info
.ignore_protect_paths
= true;
2354 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
2356 r
= setup_namespace(root_dir
, root_image
,
2357 &ns_info
, context
->read_write_paths
,
2358 needs_sandboxing
? context
->read_only_paths
: NULL
,
2359 needs_sandboxing
? context
->inaccessible_paths
: NULL
,
2363 context
->temporary_filesystems
,
2364 context
->n_temporary_filesystems
,
2367 needs_sandboxing
? context
->protect_home
: PROTECT_HOME_NO
,
2368 needs_sandboxing
? context
->protect_system
: PROTECT_SYSTEM_NO
,
2369 context
->mount_flags
,
2370 DISSECT_IMAGE_DISCARD_ON_LOOP
);
2372 bind_mount_free_many(bind_mounts
, n_bind_mounts
);
2374 /* If we couldn't set up the namespace this is probably due to a
2375 * missing capability. In this case, silently proceeed. */
2376 if (IN_SET(r
, -EPERM
, -EACCES
)) {
2377 log_unit_debug_errno(u
, r
, "Failed to set up namespace, assuming containerized execution, ignoring: %m");
2384 static int apply_working_directory(
2385 const ExecContext
*context
,
2386 const ExecParameters
*params
,
2388 const bool needs_mount_ns
,
2394 assert(exit_status
);
2396 if (context
->working_directory_home
) {
2399 *exit_status
= EXIT_CHDIR
;
2405 } else if (context
->working_directory
)
2406 wd
= context
->working_directory
;
2410 if (params
->flags
& EXEC_APPLY_CHROOT
) {
2411 if (!needs_mount_ns
&& context
->root_directory
)
2412 if (chroot(context
->root_directory
) < 0) {
2413 *exit_status
= EXIT_CHROOT
;
2419 d
= prefix_roota(context
->root_directory
, wd
);
2421 if (chdir(d
) < 0 && !context
->working_directory_missing_ok
) {
2422 *exit_status
= EXIT_CHDIR
;
2429 static int setup_keyring(
2431 const ExecContext
*context
,
2432 const ExecParameters
*p
,
2433 uid_t uid
, gid_t gid
) {
2435 key_serial_t keyring
;
2444 /* Let's set up a new per-service "session" kernel keyring for each system service. This has the benefit that
2445 * each service runs with its own keyring shared among all processes of the service, but with no hook-up beyond
2446 * that scope, and in particular no link to the per-UID keyring. If we don't do this the keyring will be
2447 * automatically created on-demand and then linked to the per-UID keyring, by the kernel. The kernel's built-in
2448 * on-demand behaviour is very appropriate for login users, but probably not so much for system services, where
2449 * UIDs are not necessarily specific to a service but reused (at least in the case of UID 0). */
2451 if (!(p
->flags
& EXEC_NEW_KEYRING
))
2454 if (context
->keyring_mode
== EXEC_KEYRING_INHERIT
)
2457 /* Acquiring a reference to the user keyring is nasty. We briefly change identity in order to get things set up
2458 * properly by the kernel. If we don't do that then we can't create it atomically, and that sucks for parallel
2459 * execution. This mimics what pam_keyinit does, too. Setting up session keyring, to be owned by the right user
2460 * & group is just as nasty as acquiring a reference to the user keyring. */
2462 saved_uid
= getuid();
2463 saved_gid
= getgid();
2465 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2466 if (setregid(gid
, -1) < 0)
2467 return log_unit_error_errno(u
, errno
, "Failed to change GID for user keyring: %m");
2470 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2471 if (setreuid(uid
, -1) < 0) {
2472 r
= log_unit_error_errno(u
, errno
, "Failed to change UID for user keyring: %m");
2477 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, 0, 0, 0, 0);
2478 if (keyring
== -1) {
2479 if (errno
== ENOSYS
)
2480 log_unit_debug_errno(u
, errno
, "Kernel keyring not supported, ignoring.");
2481 else if (IN_SET(errno
, EACCES
, EPERM
))
2482 log_unit_debug_errno(u
, errno
, "Kernel keyring access prohibited, ignoring.");
2483 else if (errno
== EDQUOT
)
2484 log_unit_debug_errno(u
, errno
, "Out of kernel keyrings to allocate, ignoring.");
2486 r
= log_unit_error_errno(u
, errno
, "Setting up kernel keyring failed: %m");
2491 /* When requested link the user keyring into the session keyring. */
2492 if (context
->keyring_mode
== EXEC_KEYRING_SHARED
) {
2494 if (keyctl(KEYCTL_LINK
,
2495 KEY_SPEC_USER_KEYRING
,
2496 KEY_SPEC_SESSION_KEYRING
, 0, 0) < 0) {
2497 r
= log_unit_error_errno(u
, errno
, "Failed to link user keyring into session keyring: %m");
2502 /* Restore uid/gid back */
2503 if (uid_is_valid(uid
) && uid
!= saved_uid
) {
2504 if (setreuid(saved_uid
, -1) < 0) {
2505 r
= log_unit_error_errno(u
, errno
, "Failed to change UID back for user keyring: %m");
2510 if (gid_is_valid(gid
) && gid
!= saved_gid
) {
2511 if (setregid(saved_gid
, -1) < 0)
2512 return log_unit_error_errno(u
, errno
, "Failed to change GID back for user keyring: %m");
2515 /* Populate they keyring with the invocation ID by default, as original saved_uid. */
2516 if (!sd_id128_is_null(u
->invocation_id
)) {
2519 key
= add_key("user", "invocation_id", &u
->invocation_id
, sizeof(u
->invocation_id
), KEY_SPEC_SESSION_KEYRING
);
2521 log_unit_debug_errno(u
, errno
, "Failed to add invocation ID to keyring, ignoring: %m");
2523 if (keyctl(KEYCTL_SETPERM
, key
,
2524 KEY_POS_VIEW
|KEY_POS_READ
|KEY_POS_SEARCH
|
2525 KEY_USR_VIEW
|KEY_USR_READ
|KEY_USR_SEARCH
, 0, 0) < 0)
2526 r
= log_unit_error_errno(u
, errno
, "Failed to restrict invocation ID permission: %m");
2531 /* Revert back uid & gid for the the last time, and exit */
2532 /* no extra logging, as only the first already reported error matters */
2533 if (getuid() != saved_uid
)
2534 (void) setreuid(saved_uid
, -1);
2536 if (getgid() != saved_gid
)
2537 (void) setregid(saved_gid
, -1);
2542 static void append_socket_pair(int *array
, unsigned *n
, const int pair
[2]) {
2550 array
[(*n
)++] = pair
[0];
2552 array
[(*n
)++] = pair
[1];
2555 static int close_remaining_fds(
2556 const ExecParameters
*params
,
2557 const ExecRuntime
*runtime
,
2558 const DynamicCreds
*dcreds
,
2561 int *fds
, unsigned n_fds
) {
2563 unsigned n_dont_close
= 0;
2564 int dont_close
[n_fds
+ 12];
2568 if (params
->stdin_fd
>= 0)
2569 dont_close
[n_dont_close
++] = params
->stdin_fd
;
2570 if (params
->stdout_fd
>= 0)
2571 dont_close
[n_dont_close
++] = params
->stdout_fd
;
2572 if (params
->stderr_fd
>= 0)
2573 dont_close
[n_dont_close
++] = params
->stderr_fd
;
2576 dont_close
[n_dont_close
++] = socket_fd
;
2578 memcpy(dont_close
+ n_dont_close
, fds
, sizeof(int) * n_fds
);
2579 n_dont_close
+= n_fds
;
2583 append_socket_pair(dont_close
, &n_dont_close
, runtime
->netns_storage_socket
);
2587 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->user
->storage_socket
);
2589 append_socket_pair(dont_close
, &n_dont_close
, dcreds
->group
->storage_socket
);
2592 if (user_lookup_fd
>= 0)
2593 dont_close
[n_dont_close
++] = user_lookup_fd
;
2595 return close_all_fds(dont_close
, n_dont_close
);
2598 static int send_user_lookup(
2606 /* Send the resolved UID/GID to PID 1 after we learnt it. We send a single datagram, containing the UID/GID
2607 * data as well as the unit name. Note that we suppress sending this if no user/group to resolve was
2610 if (user_lookup_fd
< 0)
2613 if (!uid_is_valid(uid
) && !gid_is_valid(gid
))
2616 if (writev(user_lookup_fd
,
2618 IOVEC_INIT(&uid
, sizeof(uid
)),
2619 IOVEC_INIT(&gid
, sizeof(gid
)),
2620 IOVEC_INIT_STRING(unit
->id
) }, 3) < 0)
2626 static int acquire_home(const ExecContext
*c
, uid_t uid
, const char** home
, char **buf
) {
2633 /* If WorkingDirectory=~ is set, try to acquire a usable home directory. */
2638 if (!c
->working_directory_home
)
2642 /* Hardcode /root as home directory for UID 0 */
2647 r
= get_home_dir(buf
);
2655 static int compile_suggested_paths(const ExecContext
*c
, const ExecParameters
*p
, char ***ret
) {
2656 _cleanup_strv_free_
char ** list
= NULL
;
2657 ExecDirectoryType t
;
2664 assert(c
->dynamic_user
);
2666 /* Compile a list of paths that it might make sense to read the owning UID from to use as initial candidate for
2667 * dynamic UID allocation, in order to save us from doing costly recursive chown()s of the special
2670 for (t
= 0; t
< _EXEC_DIRECTORY_TYPE_MAX
; t
++) {
2673 if (t
== EXEC_DIRECTORY_CONFIGURATION
)
2679 STRV_FOREACH(i
, c
->directories
[t
].paths
) {
2682 if (t
== EXEC_DIRECTORY_RUNTIME
)
2683 e
= strjoin(p
->prefix
[t
], "/", *i
);
2685 e
= strjoin(p
->prefix
[t
], "/private/", *i
);
2689 r
= strv_consume(&list
, e
);
2695 *ret
= TAKE_PTR(list
);
2700 static char *exec_command_line(char **argv
);
2702 static int exec_child(
2704 const ExecCommand
*command
,
2705 const ExecContext
*context
,
2706 const ExecParameters
*params
,
2707 ExecRuntime
*runtime
,
2708 DynamicCreds
*dcreds
,
2713 unsigned n_storage_fds
,
2714 unsigned n_socket_fds
,
2719 _cleanup_strv_free_
char **our_env
= NULL
, **pass_env
= NULL
, **accum_env
= NULL
, **final_argv
= NULL
;
2720 _cleanup_free_
char *home_buffer
= NULL
;
2721 _cleanup_free_ gid_t
*supplementary_gids
= NULL
;
2722 const char *username
= NULL
, *groupname
= NULL
;
2723 const char *home
= NULL
, *shell
= NULL
;
2724 dev_t journal_stream_dev
= 0;
2725 ino_t journal_stream_ino
= 0;
2726 bool needs_sandboxing
, /* Do we need to set up full sandboxing? (i.e. all namespacing, all MAC stuff, caps, yadda yadda */
2727 needs_setuid
, /* Do we need to do the actual setresuid()/setresgid() calls? */
2728 needs_mount_namespace
, /* Do we need to set up a mount namespace for this kernel? */
2729 needs_ambient_hack
; /* Do we need to apply the ambient capabilities hack? */
2731 _cleanup_free_
char *mac_selinux_context_net
= NULL
;
2732 bool use_selinux
= false;
2735 bool use_smack
= false;
2738 bool use_apparmor
= false;
2740 uid_t uid
= UID_INVALID
;
2741 gid_t gid
= GID_INVALID
;
2742 int i
, r
, ngids
= 0;
2744 ExecDirectoryType dt
;
2751 assert(exit_status
);
2753 rename_process_from_path(command
->path
);
2755 /* We reset exactly these signals, since they are the
2756 * only ones we set to SIG_IGN in the main daemon. All
2757 * others we leave untouched because we set them to
2758 * SIG_DFL or a valid handler initially, both of which
2759 * will be demoted to SIG_DFL. */
2760 (void) default_signals(SIGNALS_CRASH_HANDLER
,
2761 SIGNALS_IGNORE
, -1);
2763 if (context
->ignore_sigpipe
)
2764 (void) ignore_signals(SIGPIPE
, -1);
2766 r
= reset_signal_mask();
2768 *exit_status
= EXIT_SIGNAL_MASK
;
2769 return log_unit_error_errno(unit
, r
, "Failed to set process signal mask: %m");
2772 if (params
->idle_pipe
)
2773 do_idle_pipe_dance(params
->idle_pipe
);
2775 /* Close fds we don't need very early to make sure we don't block init reexecution because it cannot bind its
2776 * sockets. Among the fds we close are the logging fds, and we want to keep them closed, so that we don't have
2777 * any fds open we don't really want open during the transition. In order to make logging work, we switch the
2778 * log subsystem into open_when_needed mode, so that it reopens the logs on every single log call. */
2781 log_set_open_when_needed(true);
2783 /* In case anything used libc syslog(), close this here, too */
2786 n_fds
= n_storage_fds
+ n_socket_fds
;
2787 r
= close_remaining_fds(params
, runtime
, dcreds
, user_lookup_fd
, socket_fd
, fds
, n_fds
);
2789 *exit_status
= EXIT_FDS
;
2790 return log_unit_error_errno(unit
, r
, "Failed to close unwanted file descriptors: %m");
2793 if (!context
->same_pgrp
)
2795 *exit_status
= EXIT_SETSID
;
2796 return log_unit_error_errno(unit
, errno
, "Failed to create new process session: %m");
2799 exec_context_tty_reset(context
, params
);
2801 if (unit_shall_confirm_spawn(unit
)) {
2802 const char *vc
= params
->confirm_spawn
;
2803 _cleanup_free_
char *cmdline
= NULL
;
2805 cmdline
= exec_command_line(argv
);
2807 *exit_status
= EXIT_MEMORY
;
2811 r
= ask_for_confirmation(vc
, unit
, cmdline
);
2812 if (r
!= CONFIRM_EXECUTE
) {
2813 if (r
== CONFIRM_PRETEND_SUCCESS
) {
2814 *exit_status
= EXIT_SUCCESS
;
2817 *exit_status
= EXIT_CONFIRM
;
2818 log_unit_error(unit
, "Execution cancelled by the user");
2823 if (context
->dynamic_user
&& dcreds
) {
2824 _cleanup_strv_free_
char **suggested_paths
= NULL
;
2826 /* Make sure we bypass our own NSS module for any NSS checks */
2827 if (putenv((char*) "SYSTEMD_NSS_DYNAMIC_BYPASS=1") != 0) {
2828 *exit_status
= EXIT_USER
;
2829 return log_unit_error_errno(unit
, errno
, "Failed to update environment: %m");
2832 r
= compile_suggested_paths(context
, params
, &suggested_paths
);
2834 *exit_status
= EXIT_MEMORY
;
2838 r
= dynamic_creds_realize(dcreds
, suggested_paths
, &uid
, &gid
);
2840 *exit_status
= EXIT_USER
;
2842 log_unit_error(unit
, "Failed to update dynamic user credentials: User or group with specified name already exists.");
2845 return log_unit_error_errno(unit
, r
, "Failed to update dynamic user credentials: %m");
2848 if (!uid_is_valid(uid
)) {
2849 *exit_status
= EXIT_USER
;
2850 log_unit_error(unit
, "UID validation failed for \""UID_FMT
"\"", uid
);
2854 if (!gid_is_valid(gid
)) {
2855 *exit_status
= EXIT_USER
;
2856 log_unit_error(unit
, "GID validation failed for \""GID_FMT
"\"", gid
);
2861 username
= dcreds
->user
->name
;
2864 r
= get_fixed_user(context
, &username
, &uid
, &gid
, &home
, &shell
);
2866 *exit_status
= EXIT_USER
;
2867 return log_unit_error_errno(unit
, r
, "Failed to determine user credentials: %m");
2870 r
= get_fixed_group(context
, &groupname
, &gid
);
2872 *exit_status
= EXIT_GROUP
;
2873 return log_unit_error_errno(unit
, r
, "Failed to determine group credentials: %m");
2877 /* Initialize user supplementary groups and get SupplementaryGroups= ones */
2878 r
= get_supplementary_groups(context
, username
, groupname
, gid
,
2879 &supplementary_gids
, &ngids
);
2881 *exit_status
= EXIT_GROUP
;
2882 return log_unit_error_errno(unit
, r
, "Failed to determine supplementary groups: %m");
2885 r
= send_user_lookup(unit
, user_lookup_fd
, uid
, gid
);
2887 *exit_status
= EXIT_USER
;
2888 return log_unit_error_errno(unit
, r
, "Failed to send user credentials to PID1: %m");
2891 user_lookup_fd
= safe_close(user_lookup_fd
);
2893 r
= acquire_home(context
, uid
, &home
, &home_buffer
);
2895 *exit_status
= EXIT_CHDIR
;
2896 return log_unit_error_errno(unit
, r
, "Failed to determine $HOME for user: %m");
2899 /* If a socket is connected to STDIN/STDOUT/STDERR, we
2900 * must sure to drop O_NONBLOCK */
2902 (void) fd_nonblock(socket_fd
, false);
2904 r
= setup_input(context
, params
, socket_fd
, named_iofds
);
2906 *exit_status
= EXIT_STDIN
;
2907 return log_unit_error_errno(unit
, r
, "Failed to set up standard input: %m");
2910 r
= setup_output(unit
, context
, params
, STDOUT_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2912 *exit_status
= EXIT_STDOUT
;
2913 return log_unit_error_errno(unit
, r
, "Failed to set up standard output: %m");
2916 r
= setup_output(unit
, context
, params
, STDERR_FILENO
, socket_fd
, named_iofds
, basename(command
->path
), uid
, gid
, &journal_stream_dev
, &journal_stream_ino
);
2918 *exit_status
= EXIT_STDERR
;
2919 return log_unit_error_errno(unit
, r
, "Failed to set up standard error output: %m");
2922 if (params
->cgroup_path
) {
2923 r
= cg_attach_everywhere(params
->cgroup_supported
, params
->cgroup_path
, 0, NULL
, NULL
);
2925 *exit_status
= EXIT_CGROUP
;
2926 return log_unit_error_errno(unit
, r
, "Failed to attach to cgroup %s: %m", params
->cgroup_path
);
2930 if (context
->oom_score_adjust_set
) {
2931 char t
[DECIMAL_STR_MAX(context
->oom_score_adjust
)];
2933 /* When we can't make this change due to EPERM, then
2934 * let's silently skip over it. User namespaces
2935 * prohibit write access to this file, and we
2936 * shouldn't trip up over that. */
2938 sprintf(t
, "%i", context
->oom_score_adjust
);
2939 r
= write_string_file("/proc/self/oom_score_adj", t
, 0);
2940 if (IN_SET(r
, -EPERM
, -EACCES
))
2941 log_unit_debug_errno(unit
, r
, "Failed to adjust OOM setting, assuming containerized execution, ignoring: %m");
2943 *exit_status
= EXIT_OOM_ADJUST
;
2944 return log_unit_error_errno(unit
, r
, "Failed to adjust OOM setting: %m");
2948 if (context
->nice_set
)
2949 if (setpriority(PRIO_PROCESS
, 0, context
->nice
) < 0) {
2950 *exit_status
= EXIT_NICE
;
2951 return log_unit_error_errno(unit
, errno
, "Failed to set up process scheduling priority (nice level): %m");
2954 if (context
->cpu_sched_set
) {
2955 struct sched_param param
= {
2956 .sched_priority
= context
->cpu_sched_priority
,
2959 r
= sched_setscheduler(0,
2960 context
->cpu_sched_policy
|
2961 (context
->cpu_sched_reset_on_fork
?
2962 SCHED_RESET_ON_FORK
: 0),
2965 *exit_status
= EXIT_SETSCHEDULER
;
2966 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU scheduling: %m");
2970 if (context
->cpuset
)
2971 if (sched_setaffinity(0, CPU_ALLOC_SIZE(context
->cpuset_ncpus
), context
->cpuset
) < 0) {
2972 *exit_status
= EXIT_CPUAFFINITY
;
2973 return log_unit_error_errno(unit
, errno
, "Failed to set up CPU affinity: %m");
2976 if (context
->ioprio_set
)
2977 if (ioprio_set(IOPRIO_WHO_PROCESS
, 0, context
->ioprio
) < 0) {
2978 *exit_status
= EXIT_IOPRIO
;
2979 return log_unit_error_errno(unit
, errno
, "Failed to set up IO scheduling priority: %m");
2982 if (context
->timer_slack_nsec
!= NSEC_INFINITY
)
2983 if (prctl(PR_SET_TIMERSLACK
, context
->timer_slack_nsec
) < 0) {
2984 *exit_status
= EXIT_TIMERSLACK
;
2985 return log_unit_error_errno(unit
, errno
, "Failed to set up timer slack: %m");
2988 if (context
->personality
!= PERSONALITY_INVALID
) {
2989 r
= safe_personality(context
->personality
);
2991 *exit_status
= EXIT_PERSONALITY
;
2992 return log_unit_error_errno(unit
, r
, "Failed to set up execution domain (personality): %m");
2996 if (context
->utmp_id
)
2997 utmp_put_init_process(context
->utmp_id
, getpid_cached(), getsid(0),
2999 context
->utmp_mode
== EXEC_UTMP_INIT
? INIT_PROCESS
:
3000 context
->utmp_mode
== EXEC_UTMP_LOGIN
? LOGIN_PROCESS
:
3004 if (context
->user
) {
3005 r
= chown_terminal(STDIN_FILENO
, uid
);
3007 *exit_status
= EXIT_STDIN
;
3008 return log_unit_error_errno(unit
, r
, "Failed to change ownership of terminal: %m");
3012 /* If delegation is enabled we'll pass ownership of the cgroup to the user of the new process. On cgroupsv1
3013 * this is only about systemd's own hierarchy, i.e. not the controller hierarchies, simply because that's not
3014 * safe. On cgroupsv2 there's only one hierarchy anyway, and delegation is safe there, hence in that case only
3015 * touch a single hierarchy too. */
3016 if (params
->cgroup_path
&& context
->user
&& (params
->flags
& EXEC_CGROUP_DELEGATE
)) {
3017 r
= cg_set_access(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, uid
, gid
);
3019 *exit_status
= EXIT_CGROUP
;
3020 return log_unit_error_errno(unit
, r
, "Failed to adjust control group access: %m");
3024 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3025 r
= setup_exec_directory(context
, params
, uid
, gid
, dt
, exit_status
);
3027 return log_unit_error_errno(unit
, r
, "Failed to set up special execution directory in %s: %m", params
->prefix
[dt
]);
3030 r
= build_environment(
3042 *exit_status
= EXIT_MEMORY
;
3046 r
= build_pass_environment(context
, &pass_env
);
3048 *exit_status
= EXIT_MEMORY
;
3052 accum_env
= strv_env_merge(5,
3053 params
->environment
,
3056 context
->environment
,
3060 *exit_status
= EXIT_MEMORY
;
3063 accum_env
= strv_env_clean(accum_env
);
3065 (void) umask(context
->umask
);
3067 r
= setup_keyring(unit
, context
, params
, uid
, gid
);
3069 *exit_status
= EXIT_KEYRING
;
3070 return log_unit_error_errno(unit
, r
, "Failed to set up kernel keyring: %m");
3073 /* We need sandboxing if the caller asked us to apply it and the command isn't explicitly excepted from it */
3074 needs_sandboxing
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& EXEC_COMMAND_FULLY_PRIVILEGED
);
3076 /* We need the ambient capability hack, if the caller asked us to apply it and the command is marked for it, and the kernel doesn't actually support ambient caps */
3077 needs_ambient_hack
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && (command
->flags
& EXEC_COMMAND_AMBIENT_MAGIC
) && !ambient_capabilities_supported();
3079 /* We need setresuid() if the caller asked us to apply sandboxing and the command isn't explicitly excepted from either whole sandboxing or just setresuid() itself, and the ambient hack is not desired */
3080 if (needs_ambient_hack
)
3081 needs_setuid
= false;
3083 needs_setuid
= (params
->flags
& EXEC_APPLY_SANDBOXING
) && !(command
->flags
& (EXEC_COMMAND_FULLY_PRIVILEGED
|EXEC_COMMAND_NO_SETUID
));
3085 if (needs_sandboxing
) {
3086 /* MAC enablement checks need to be done before a new mount ns is created, as they rely on /sys being
3087 * present. The actual MAC context application will happen later, as late as possible, to avoid
3088 * impacting our own code paths. */
3091 use_selinux
= mac_selinux_use();
3094 use_smack
= mac_smack_use();
3097 use_apparmor
= mac_apparmor_use();
3102 if (context
->pam_name
&& username
) {
3103 r
= setup_pam(context
->pam_name
, username
, uid
, gid
, context
->tty_path
, &accum_env
, fds
, n_fds
);
3105 *exit_status
= EXIT_PAM
;
3106 return log_unit_error_errno(unit
, r
, "Failed to set up PAM session: %m");
3111 if (context
->private_network
&& runtime
&& runtime
->netns_storage_socket
[0] >= 0) {
3112 if (ns_type_supported(NAMESPACE_NET
)) {
3113 r
= setup_netns(runtime
->netns_storage_socket
);
3115 *exit_status
= EXIT_NETWORK
;
3116 return log_unit_error_errno(unit
, r
, "Failed to set up network namespacing: %m");
3119 log_unit_warning(unit
, "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
3122 needs_mount_namespace
= exec_needs_mount_namespace(context
, params
, runtime
);
3123 if (needs_mount_namespace
) {
3124 r
= apply_mount_namespace(unit
, command
, context
, params
, runtime
);
3126 *exit_status
= EXIT_NAMESPACE
;
3127 return log_unit_error_errno(unit
, r
, "Failed to set up mount namespacing: %m");
3131 /* Apply just after mount namespace setup */
3132 r
= apply_working_directory(context
, params
, home
, needs_mount_namespace
, exit_status
);
3134 return log_unit_error_errno(unit
, r
, "Changing to the requested working directory failed: %m");
3136 /* Drop groups as early as possbile */
3138 r
= enforce_groups(gid
, supplementary_gids
, ngids
);
3140 *exit_status
= EXIT_GROUP
;
3141 return log_unit_error_errno(unit
, r
, "Changing group credentials failed: %m");
3145 if (needs_sandboxing
) {
3147 if (use_selinux
&& params
->selinux_context_net
&& socket_fd
>= 0) {
3148 r
= mac_selinux_get_child_mls_label(socket_fd
, command
->path
, context
->selinux_context
, &mac_selinux_context_net
);
3150 *exit_status
= EXIT_SELINUX_CONTEXT
;
3151 return log_unit_error_errno(unit
, r
, "Failed to determine SELinux context: %m");
3156 if (context
->private_users
) {
3157 r
= setup_private_users(uid
, gid
);
3159 *exit_status
= EXIT_USER
;
3160 return log_unit_error_errno(unit
, r
, "Failed to set up user namespacing: %m");
3165 /* We repeat the fd closing here, to make sure that nothing is leaked from the PAM modules. Note that we are
3166 * more aggressive this time since socket_fd and the netns fds we don't need anymore. The custom endpoint fd
3167 * was needed to upload the policy and can now be closed as well. */
3168 r
= close_all_fds(fds
, n_fds
);
3170 r
= shift_fds(fds
, n_fds
);
3172 r
= flags_fds(fds
, n_storage_fds
, n_socket_fds
, context
->non_blocking
);
3174 *exit_status
= EXIT_FDS
;
3175 return log_unit_error_errno(unit
, r
, "Failed to adjust passed file descriptors: %m");
3178 secure_bits
= context
->secure_bits
;
3180 if (needs_sandboxing
) {
3183 for (i
= 0; i
< _RLIMIT_MAX
; i
++) {
3185 if (!context
->rlimit
[i
])
3188 r
= setrlimit_closest(i
, context
->rlimit
[i
]);
3190 *exit_status
= EXIT_LIMITS
;
3191 return log_unit_error_errno(unit
, r
, "Failed to adjust resource limit %s: %m", rlimit_to_string(i
));
3195 /* Set the RTPRIO resource limit to 0, but only if nothing else was explicitly requested. */
3196 if (context
->restrict_realtime
&& !context
->rlimit
[RLIMIT_RTPRIO
]) {
3197 if (setrlimit(RLIMIT_RTPRIO
, &RLIMIT_MAKE_CONST(0)) < 0) {
3198 *exit_status
= EXIT_LIMITS
;
3199 return log_unit_error_errno(unit
, errno
, "Failed to adjust RLIMIT_RTPRIO resource limit: %m");
3204 /* LSM Smack needs the capability CAP_MAC_ADMIN to change the current execution security context of the
3205 * process. This is the latest place before dropping capabilities. Other MAC context are set later. */
3207 r
= setup_smack(context
, command
);
3209 *exit_status
= EXIT_SMACK_PROCESS_LABEL
;
3210 return log_unit_error_errno(unit
, r
, "Failed to set SMACK process label: %m");
3215 bset
= context
->capability_bounding_set
;
3216 /* If the ambient caps hack is enabled (which means the kernel can't do them, and the user asked for
3217 * our magic fallback), then let's add some extra caps, so that the service can drop privs of its own,
3218 * instead of us doing that */
3219 if (needs_ambient_hack
)
3220 bset
|= (UINT64_C(1) << CAP_SETPCAP
) |
3221 (UINT64_C(1) << CAP_SETUID
) |
3222 (UINT64_C(1) << CAP_SETGID
);
3224 if (!cap_test_all(bset
)) {
3225 r
= capability_bounding_set_drop(bset
, false);
3227 *exit_status
= EXIT_CAPABILITIES
;
3228 return log_unit_error_errno(unit
, r
, "Failed to drop capabilities: %m");
3232 /* This is done before enforce_user, but ambient set
3233 * does not survive over setresuid() if keep_caps is not set. */
3234 if (!needs_ambient_hack
&&
3235 context
->capability_ambient_set
!= 0) {
3236 r
= capability_ambient_set_apply(context
->capability_ambient_set
, true);
3238 *exit_status
= EXIT_CAPABILITIES
;
3239 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (before UID change): %m");
3245 if (context
->user
) {
3246 r
= enforce_user(context
, uid
);
3248 *exit_status
= EXIT_USER
;
3249 return log_unit_error_errno(unit
, r
, "Failed to change UID to " UID_FMT
": %m", uid
);
3252 if (!needs_ambient_hack
&&
3253 context
->capability_ambient_set
!= 0) {
3255 /* Fix the ambient capabilities after user change. */
3256 r
= capability_ambient_set_apply(context
->capability_ambient_set
, false);
3258 *exit_status
= EXIT_CAPABILITIES
;
3259 return log_unit_error_errno(unit
, r
, "Failed to apply ambient capabilities (after UID change): %m");
3262 /* If we were asked to change user and ambient capabilities
3263 * were requested, we had to add keep-caps to the securebits
3264 * so that we would maintain the inherited capability set
3265 * through the setresuid(). Make sure that the bit is added
3266 * also to the context secure_bits so that we don't try to
3267 * drop the bit away next. */
3269 secure_bits
|= 1<<SECURE_KEEP_CAPS
;
3274 if (needs_sandboxing
) {
3275 /* Apply other MAC contexts late, but before seccomp syscall filtering, as those should really be last to
3276 * influence our own codepaths as little as possible. Moreover, applying MAC contexts usually requires
3277 * syscalls that are subject to seccomp filtering, hence should probably be applied before the syscalls
3278 * are restricted. */
3282 char *exec_context
= mac_selinux_context_net
?: context
->selinux_context
;
3285 r
= setexeccon(exec_context
);
3287 *exit_status
= EXIT_SELINUX_CONTEXT
;
3288 return log_unit_error_errno(unit
, r
, "Failed to change SELinux context to %s: %m", exec_context
);
3295 if (use_apparmor
&& context
->apparmor_profile
) {
3296 r
= aa_change_onexec(context
->apparmor_profile
);
3297 if (r
< 0 && !context
->apparmor_profile_ignore
) {
3298 *exit_status
= EXIT_APPARMOR_PROFILE
;
3299 return log_unit_error_errno(unit
, errno
, "Failed to prepare AppArmor profile change to %s: %m", context
->apparmor_profile
);
3304 /* PR_GET_SECUREBITS is not privileged, while PR_SET_SECUREBITS is. So to suppress potential EPERMs
3305 * we'll try not to call PR_SET_SECUREBITS unless necessary. */
3306 if (prctl(PR_GET_SECUREBITS
) != secure_bits
)
3307 if (prctl(PR_SET_SECUREBITS
, secure_bits
) < 0) {
3308 *exit_status
= EXIT_SECUREBITS
;
3309 return log_unit_error_errno(unit
, errno
, "Failed to set process secure bits: %m");
3312 if (context_has_no_new_privileges(context
))
3313 if (prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0) < 0) {
3314 *exit_status
= EXIT_NO_NEW_PRIVILEGES
;
3315 return log_unit_error_errno(unit
, errno
, "Failed to disable new privileges: %m");
3319 r
= apply_address_families(unit
, context
);
3321 *exit_status
= EXIT_ADDRESS_FAMILIES
;
3322 return log_unit_error_errno(unit
, r
, "Failed to restrict address families: %m");
3325 r
= apply_memory_deny_write_execute(unit
, context
);
3327 *exit_status
= EXIT_SECCOMP
;
3328 return log_unit_error_errno(unit
, r
, "Failed to disable writing to executable memory: %m");
3331 r
= apply_restrict_realtime(unit
, context
);
3333 *exit_status
= EXIT_SECCOMP
;
3334 return log_unit_error_errno(unit
, r
, "Failed to apply realtime restrictions: %m");
3337 r
= apply_restrict_namespaces(unit
, context
);
3339 *exit_status
= EXIT_SECCOMP
;
3340 return log_unit_error_errno(unit
, r
, "Failed to apply namespace restrictions: %m");
3343 r
= apply_protect_sysctl(unit
, context
);
3345 *exit_status
= EXIT_SECCOMP
;
3346 return log_unit_error_errno(unit
, r
, "Failed to apply sysctl restrictions: %m");
3349 r
= apply_protect_kernel_modules(unit
, context
);
3351 *exit_status
= EXIT_SECCOMP
;
3352 return log_unit_error_errno(unit
, r
, "Failed to apply module loading restrictions: %m");
3355 r
= apply_private_devices(unit
, context
);
3357 *exit_status
= EXIT_SECCOMP
;
3358 return log_unit_error_errno(unit
, r
, "Failed to set up private devices: %m");
3361 r
= apply_syscall_archs(unit
, context
);
3363 *exit_status
= EXIT_SECCOMP
;
3364 return log_unit_error_errno(unit
, r
, "Failed to apply syscall architecture restrictions: %m");
3367 r
= apply_lock_personality(unit
, context
);
3369 *exit_status
= EXIT_SECCOMP
;
3370 return log_unit_error_errno(unit
, r
, "Failed to lock personalities: %m");
3373 /* This really should remain the last step before the execve(), to make sure our own code is unaffected
3374 * by the filter as little as possible. */
3375 r
= apply_syscall_filter(unit
, context
, needs_ambient_hack
);
3377 *exit_status
= EXIT_SECCOMP
;
3378 return log_unit_error_errno(unit
, r
, "Failed to apply system call filters: %m");
3383 if (!strv_isempty(context
->unset_environment
)) {
3386 ee
= strv_env_delete(accum_env
, 1, context
->unset_environment
);
3388 *exit_status
= EXIT_MEMORY
;
3392 strv_free(accum_env
);
3396 final_argv
= replace_env_argv(argv
, accum_env
);
3398 *exit_status
= EXIT_MEMORY
;
3402 if (DEBUG_LOGGING
) {
3403 _cleanup_free_
char *line
;
3405 line
= exec_command_line(final_argv
);
3407 log_struct(LOG_DEBUG
,
3408 "EXECUTABLE=%s", command
->path
,
3409 LOG_UNIT_MESSAGE(unit
, "Executing: %s", line
),
3411 LOG_UNIT_INVOCATION_ID(unit
),
3416 execve(command
->path
, final_argv
, accum_env
);
3418 if (errno
== ENOENT
&& (command
->flags
& EXEC_COMMAND_IGNORE_FAILURE
)) {
3420 log_struct_errno(LOG_INFO
, errno
,
3421 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3423 LOG_UNIT_INVOCATION_ID(unit
),
3424 LOG_UNIT_MESSAGE(unit
, "Executable %s missing, skipping: %m",
3426 "EXECUTABLE=%s", command
->path
,
3432 *exit_status
= EXIT_EXEC
;
3433 return log_unit_error_errno(unit
, errno
, "Failed to execute command: %m");
3436 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
);
3437 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]);
3439 int exec_spawn(Unit
*unit
,
3440 ExecCommand
*command
,
3441 const ExecContext
*context
,
3442 const ExecParameters
*params
,
3443 ExecRuntime
*runtime
,
3444 DynamicCreds
*dcreds
,
3447 _cleanup_strv_free_
char **files_env
= NULL
;
3449 unsigned n_storage_fds
= 0, n_socket_fds
= 0;
3450 _cleanup_free_
char *line
= NULL
;
3452 int named_iofds
[3] = { -1, -1, -1 };
3461 assert(params
->fds
|| (params
->n_storage_fds
+ params
->n_socket_fds
<= 0));
3463 if (context
->std_input
== EXEC_INPUT_SOCKET
||
3464 context
->std_output
== EXEC_OUTPUT_SOCKET
||
3465 context
->std_error
== EXEC_OUTPUT_SOCKET
) {
3467 if (params
->n_socket_fds
> 1) {
3468 log_unit_error(unit
, "Got more than one socket.");
3472 if (params
->n_socket_fds
== 0) {
3473 log_unit_error(unit
, "Got no socket.");
3477 socket_fd
= params
->fds
[0];
3481 n_storage_fds
= params
->n_storage_fds
;
3482 n_socket_fds
= params
->n_socket_fds
;
3485 r
= exec_context_named_iofds(context
, params
, named_iofds
);
3487 return log_unit_error_errno(unit
, r
, "Failed to load a named file descriptor: %m");
3489 r
= exec_context_load_environment(unit
, context
, &files_env
);
3491 return log_unit_error_errno(unit
, r
, "Failed to load environment files: %m");
3493 argv
= params
->argv
?: command
->argv
;
3494 line
= exec_command_line(argv
);
3498 log_struct(LOG_DEBUG
,
3499 LOG_UNIT_MESSAGE(unit
, "About to execute: %s", line
),
3500 "EXECUTABLE=%s", command
->path
,
3502 LOG_UNIT_INVOCATION_ID(unit
),
3507 return log_unit_error_errno(unit
, errno
, "Failed to fork: %m");
3510 int exit_status
= EXIT_SUCCESS
;
3512 r
= exec_child(unit
,
3525 unit
->manager
->user_lookup_fds
[1],
3529 log_struct_errno(LOG_ERR
, r
,
3530 "MESSAGE_ID=" SD_MESSAGE_SPAWN_FAILED_STR
,
3532 LOG_UNIT_INVOCATION_ID(unit
),
3533 LOG_UNIT_MESSAGE(unit
, "Failed at step %s spawning %s: %m",
3534 exit_status_to_string(exit_status
, EXIT_STATUS_SYSTEMD
),
3536 "EXECUTABLE=%s", command
->path
,
3543 log_unit_debug(unit
, "Forked %s as "PID_FMT
, command
->path
, pid
);
3545 /* We add the new process to the cgroup both in the child (so
3546 * that we can be sure that no user code is ever executed
3547 * outside of the cgroup) and in the parent (so that we can be
3548 * sure that when we kill the cgroup the process will be
3550 if (params
->cgroup_path
)
3551 (void) cg_attach(SYSTEMD_CGROUP_CONTROLLER
, params
->cgroup_path
, pid
);
3553 exec_status_start(&command
->exec_status
, pid
);
3559 void exec_context_init(ExecContext
*c
) {
3560 ExecDirectoryType i
;
3565 c
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 0);
3566 c
->cpu_sched_policy
= SCHED_OTHER
;
3567 c
->syslog_priority
= LOG_DAEMON
|LOG_INFO
;
3568 c
->syslog_level_prefix
= true;
3569 c
->ignore_sigpipe
= true;
3570 c
->timer_slack_nsec
= NSEC_INFINITY
;
3571 c
->personality
= PERSONALITY_INVALID
;
3572 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3573 c
->directories
[i
].mode
= 0755;
3574 c
->capability_bounding_set
= CAP_ALL
;
3575 c
->restrict_namespaces
= NAMESPACE_FLAGS_ALL
;
3576 c
->log_level_max
= -1;
3579 void exec_context_done(ExecContext
*c
) {
3580 ExecDirectoryType i
;
3585 c
->environment
= strv_free(c
->environment
);
3586 c
->environment_files
= strv_free(c
->environment_files
);
3587 c
->pass_environment
= strv_free(c
->pass_environment
);
3588 c
->unset_environment
= strv_free(c
->unset_environment
);
3590 for (l
= 0; l
< ELEMENTSOF(c
->rlimit
); l
++)
3591 c
->rlimit
[l
] = mfree(c
->rlimit
[l
]);
3593 for (l
= 0; l
< 3; l
++) {
3594 c
->stdio_fdname
[l
] = mfree(c
->stdio_fdname
[l
]);
3595 c
->stdio_file
[l
] = mfree(c
->stdio_file
[l
]);
3598 c
->working_directory
= mfree(c
->working_directory
);
3599 c
->root_directory
= mfree(c
->root_directory
);
3600 c
->root_image
= mfree(c
->root_image
);
3601 c
->tty_path
= mfree(c
->tty_path
);
3602 c
->syslog_identifier
= mfree(c
->syslog_identifier
);
3603 c
->user
= mfree(c
->user
);
3604 c
->group
= mfree(c
->group
);
3606 c
->supplementary_groups
= strv_free(c
->supplementary_groups
);
3608 c
->pam_name
= mfree(c
->pam_name
);
3610 c
->read_only_paths
= strv_free(c
->read_only_paths
);
3611 c
->read_write_paths
= strv_free(c
->read_write_paths
);
3612 c
->inaccessible_paths
= strv_free(c
->inaccessible_paths
);
3614 bind_mount_free_many(c
->bind_mounts
, c
->n_bind_mounts
);
3615 c
->bind_mounts
= NULL
;
3616 c
->n_bind_mounts
= 0;
3617 temporary_filesystem_free_many(c
->temporary_filesystems
, c
->n_temporary_filesystems
);
3618 c
->temporary_filesystems
= NULL
;
3619 c
->n_temporary_filesystems
= 0;
3621 c
->cpuset
= cpu_set_mfree(c
->cpuset
);
3623 c
->utmp_id
= mfree(c
->utmp_id
);
3624 c
->selinux_context
= mfree(c
->selinux_context
);
3625 c
->apparmor_profile
= mfree(c
->apparmor_profile
);
3626 c
->smack_process_label
= mfree(c
->smack_process_label
);
3628 c
->syscall_filter
= hashmap_free(c
->syscall_filter
);
3629 c
->syscall_archs
= set_free(c
->syscall_archs
);
3630 c
->address_families
= set_free(c
->address_families
);
3632 for (i
= 0; i
< _EXEC_DIRECTORY_TYPE_MAX
; i
++)
3633 c
->directories
[i
].paths
= strv_free(c
->directories
[i
].paths
);
3635 c
->log_level_max
= -1;
3637 exec_context_free_log_extra_fields(c
);
3639 c
->stdin_data
= mfree(c
->stdin_data
);
3640 c
->stdin_data_size
= 0;
3643 int exec_context_destroy_runtime_directory(const ExecContext
*c
, const char *runtime_prefix
) {
3648 if (!runtime_prefix
)
3651 STRV_FOREACH(i
, c
->directories
[EXEC_DIRECTORY_RUNTIME
].paths
) {
3652 _cleanup_free_
char *p
;
3654 p
= strjoin(runtime_prefix
, "/", *i
);
3658 /* We execute this synchronously, since we need to be sure this is gone when we start the service
3660 (void) rm_rf(p
, REMOVE_ROOT
);
3666 static void exec_command_done(ExecCommand
*c
) {
3669 c
->path
= mfree(c
->path
);
3671 c
->argv
= strv_free(c
->argv
);
3674 void exec_command_done_array(ExecCommand
*c
, unsigned n
) {
3677 for (i
= 0; i
< n
; i
++)
3678 exec_command_done(c
+i
);
3681 ExecCommand
* exec_command_free_list(ExecCommand
*c
) {
3685 LIST_REMOVE(command
, c
, i
);
3686 exec_command_done(i
);
3693 void exec_command_free_array(ExecCommand
**c
, unsigned n
) {
3696 for (i
= 0; i
< n
; i
++)
3697 c
[i
] = exec_command_free_list(c
[i
]);
3700 typedef struct InvalidEnvInfo
{
3705 static void invalid_env(const char *p
, void *userdata
) {
3706 InvalidEnvInfo
*info
= userdata
;
3708 log_unit_error(info
->unit
, "Ignoring invalid environment assignment '%s': %s", p
, info
->path
);
3711 const char* exec_context_fdname(const ExecContext
*c
, int fd_index
) {
3717 if (c
->std_input
!= EXEC_INPUT_NAMED_FD
)
3720 return c
->stdio_fdname
[STDIN_FILENO
] ?: "stdin";
3723 if (c
->std_output
!= EXEC_OUTPUT_NAMED_FD
)
3726 return c
->stdio_fdname
[STDOUT_FILENO
] ?: "stdout";
3729 if (c
->std_error
!= EXEC_OUTPUT_NAMED_FD
)
3732 return c
->stdio_fdname
[STDERR_FILENO
] ?: "stderr";
3739 static int exec_context_named_iofds(const ExecContext
*c
, const ExecParameters
*p
, int named_iofds
[3]) {
3740 unsigned i
, targets
;
3741 const char* stdio_fdname
[3];
3747 targets
= (c
->std_input
== EXEC_INPUT_NAMED_FD
) +
3748 (c
->std_output
== EXEC_OUTPUT_NAMED_FD
) +
3749 (c
->std_error
== EXEC_OUTPUT_NAMED_FD
);
3751 for (i
= 0; i
< 3; i
++)
3752 stdio_fdname
[i
] = exec_context_fdname(c
, i
);
3754 n_fds
= p
->n_storage_fds
+ p
->n_socket_fds
;
3756 for (i
= 0; i
< n_fds
&& targets
> 0; i
++)
3757 if (named_iofds
[STDIN_FILENO
] < 0 &&
3758 c
->std_input
== EXEC_INPUT_NAMED_FD
&&
3759 stdio_fdname
[STDIN_FILENO
] &&
3760 streq(p
->fd_names
[i
], stdio_fdname
[STDIN_FILENO
])) {
3762 named_iofds
[STDIN_FILENO
] = p
->fds
[i
];
3765 } else if (named_iofds
[STDOUT_FILENO
] < 0 &&
3766 c
->std_output
== EXEC_OUTPUT_NAMED_FD
&&
3767 stdio_fdname
[STDOUT_FILENO
] &&
3768 streq(p
->fd_names
[i
], stdio_fdname
[STDOUT_FILENO
])) {
3770 named_iofds
[STDOUT_FILENO
] = p
->fds
[i
];
3773 } else if (named_iofds
[STDERR_FILENO
] < 0 &&
3774 c
->std_error
== EXEC_OUTPUT_NAMED_FD
&&
3775 stdio_fdname
[STDERR_FILENO
] &&
3776 streq(p
->fd_names
[i
], stdio_fdname
[STDERR_FILENO
])) {
3778 named_iofds
[STDERR_FILENO
] = p
->fds
[i
];
3782 return targets
== 0 ? 0 : -ENOENT
;
3785 static int exec_context_load_environment(const Unit
*unit
, const ExecContext
*c
, char ***l
) {
3786 char **i
, **r
= NULL
;
3791 STRV_FOREACH(i
, c
->environment_files
) {
3795 bool ignore
= false;
3797 _cleanup_globfree_ glob_t pglob
= {};
3806 if (!path_is_absolute(fn
)) {
3814 /* Filename supports globbing, take all matching files */
3815 k
= safe_glob(fn
, 0, &pglob
);
3824 /* When we don't match anything, -ENOENT should be returned */
3825 assert(pglob
.gl_pathc
> 0);
3827 for (n
= 0; n
< pglob
.gl_pathc
; n
++) {
3828 k
= load_env_file(NULL
, pglob
.gl_pathv
[n
], NULL
, &p
);
3836 /* Log invalid environment variables with filename */
3838 InvalidEnvInfo info
= {
3840 .path
= pglob
.gl_pathv
[n
]
3843 p
= strv_env_clean_with_callback(p
, invalid_env
, &info
);
3851 m
= strv_env_merge(2, r
, p
);
3867 static bool tty_may_match_dev_console(const char *tty
) {
3868 _cleanup_free_
char *resolved
= NULL
;
3873 tty
= skip_dev_prefix(tty
);
3875 /* trivial identity? */
3876 if (streq(tty
, "console"))
3879 if (resolve_dev_console(&resolved
) < 0)
3880 return true; /* if we could not resolve, assume it may */
3882 /* "tty0" means the active VC, so it may be the same sometimes */
3883 return streq(resolved
, tty
) || (streq(resolved
, "tty0") && tty_is_vc(tty
));
3886 bool exec_context_may_touch_console(const ExecContext
*ec
) {
3888 return (ec
->tty_reset
||
3890 ec
->tty_vt_disallocate
||
3891 is_terminal_input(ec
->std_input
) ||
3892 is_terminal_output(ec
->std_output
) ||
3893 is_terminal_output(ec
->std_error
)) &&
3894 tty_may_match_dev_console(exec_context_tty_path(ec
));
3897 static void strv_fprintf(FILE *f
, char **l
) {
3903 fprintf(f
, " %s", *g
);
3906 void exec_context_dump(const ExecContext
*c
, FILE* f
, const char *prefix
) {
3907 ExecDirectoryType dt
;
3915 prefix
= strempty(prefix
);
3919 "%sWorkingDirectory: %s\n"
3920 "%sRootDirectory: %s\n"
3921 "%sNonBlocking: %s\n"
3922 "%sPrivateTmp: %s\n"
3923 "%sPrivateDevices: %s\n"
3924 "%sProtectKernelTunables: %s\n"
3925 "%sProtectKernelModules: %s\n"
3926 "%sProtectControlGroups: %s\n"
3927 "%sPrivateNetwork: %s\n"
3928 "%sPrivateUsers: %s\n"
3929 "%sProtectHome: %s\n"
3930 "%sProtectSystem: %s\n"
3931 "%sMountAPIVFS: %s\n"
3932 "%sIgnoreSIGPIPE: %s\n"
3933 "%sMemoryDenyWriteExecute: %s\n"
3934 "%sRestrictRealtime: %s\n"
3935 "%sKeyringMode: %s\n",
3937 prefix
, c
->working_directory
? c
->working_directory
: "/",
3938 prefix
, c
->root_directory
? c
->root_directory
: "/",
3939 prefix
, yes_no(c
->non_blocking
),
3940 prefix
, yes_no(c
->private_tmp
),
3941 prefix
, yes_no(c
->private_devices
),
3942 prefix
, yes_no(c
->protect_kernel_tunables
),
3943 prefix
, yes_no(c
->protect_kernel_modules
),
3944 prefix
, yes_no(c
->protect_control_groups
),
3945 prefix
, yes_no(c
->private_network
),
3946 prefix
, yes_no(c
->private_users
),
3947 prefix
, protect_home_to_string(c
->protect_home
),
3948 prefix
, protect_system_to_string(c
->protect_system
),
3949 prefix
, yes_no(c
->mount_apivfs
),
3950 prefix
, yes_no(c
->ignore_sigpipe
),
3951 prefix
, yes_no(c
->memory_deny_write_execute
),
3952 prefix
, yes_no(c
->restrict_realtime
),
3953 prefix
, exec_keyring_mode_to_string(c
->keyring_mode
));
3956 fprintf(f
, "%sRootImage: %s\n", prefix
, c
->root_image
);
3958 STRV_FOREACH(e
, c
->environment
)
3959 fprintf(f
, "%sEnvironment: %s\n", prefix
, *e
);
3961 STRV_FOREACH(e
, c
->environment_files
)
3962 fprintf(f
, "%sEnvironmentFile: %s\n", prefix
, *e
);
3964 STRV_FOREACH(e
, c
->pass_environment
)
3965 fprintf(f
, "%sPassEnvironment: %s\n", prefix
, *e
);
3967 STRV_FOREACH(e
, c
->unset_environment
)
3968 fprintf(f
, "%sUnsetEnvironment: %s\n", prefix
, *e
);
3970 fprintf(f
, "%sRuntimeDirectoryPreserve: %s\n", prefix
, exec_preserve_mode_to_string(c
->runtime_directory_preserve_mode
));
3972 for (dt
= 0; dt
< _EXEC_DIRECTORY_TYPE_MAX
; dt
++) {
3973 fprintf(f
, "%s%sMode: %04o\n", prefix
, exec_directory_type_to_string(dt
), c
->directories
[dt
].mode
);
3975 STRV_FOREACH(d
, c
->directories
[dt
].paths
)
3976 fprintf(f
, "%s%s: %s\n", prefix
, exec_directory_type_to_string(dt
), *d
);
3984 if (c
->oom_score_adjust_set
)
3986 "%sOOMScoreAdjust: %i\n",
3987 prefix
, c
->oom_score_adjust
);
3989 for (i
= 0; i
< RLIM_NLIMITS
; i
++)
3991 fprintf(f
, "%s%s: " RLIM_FMT
"\n",
3992 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_max
);
3993 fprintf(f
, "%s%sSoft: " RLIM_FMT
"\n",
3994 prefix
, rlimit_to_string(i
), c
->rlimit
[i
]->rlim_cur
);
3997 if (c
->ioprio_set
) {
3998 _cleanup_free_
char *class_str
= NULL
;
4000 r
= ioprio_class_to_string_alloc(IOPRIO_PRIO_CLASS(c
->ioprio
), &class_str
);
4002 fprintf(f
, "%sIOSchedulingClass: %s\n", prefix
, class_str
);
4004 fprintf(f
, "%sIOPriority: %lu\n", prefix
, IOPRIO_PRIO_DATA(c
->ioprio
));
4007 if (c
->cpu_sched_set
) {
4008 _cleanup_free_
char *policy_str
= NULL
;
4010 r
= sched_policy_to_string_alloc(c
->cpu_sched_policy
, &policy_str
);
4012 fprintf(f
, "%sCPUSchedulingPolicy: %s\n", prefix
, policy_str
);
4015 "%sCPUSchedulingPriority: %i\n"
4016 "%sCPUSchedulingResetOnFork: %s\n",
4017 prefix
, c
->cpu_sched_priority
,
4018 prefix
, yes_no(c
->cpu_sched_reset_on_fork
));
4022 fprintf(f
, "%sCPUAffinity:", prefix
);
4023 for (i
= 0; i
< c
->cpuset_ncpus
; i
++)
4024 if (CPU_ISSET_S(i
, CPU_ALLOC_SIZE(c
->cpuset_ncpus
), c
->cpuset
))
4025 fprintf(f
, " %u", i
);
4029 if (c
->timer_slack_nsec
!= NSEC_INFINITY
)
4030 fprintf(f
, "%sTimerSlackNSec: "NSEC_FMT
"\n", prefix
, c
->timer_slack_nsec
);
4033 "%sStandardInput: %s\n"
4034 "%sStandardOutput: %s\n"
4035 "%sStandardError: %s\n",
4036 prefix
, exec_input_to_string(c
->std_input
),
4037 prefix
, exec_output_to_string(c
->std_output
),
4038 prefix
, exec_output_to_string(c
->std_error
));
4040 if (c
->std_input
== EXEC_INPUT_NAMED_FD
)
4041 fprintf(f
, "%sStandardInputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDIN_FILENO
]);
4042 if (c
->std_output
== EXEC_OUTPUT_NAMED_FD
)
4043 fprintf(f
, "%sStandardOutputFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDOUT_FILENO
]);
4044 if (c
->std_error
== EXEC_OUTPUT_NAMED_FD
)
4045 fprintf(f
, "%sStandardErrorFileDescriptorName: %s\n", prefix
, c
->stdio_fdname
[STDERR_FILENO
]);
4047 if (c
->std_input
== EXEC_INPUT_FILE
)
4048 fprintf(f
, "%sStandardInputFile: %s\n", prefix
, c
->stdio_file
[STDIN_FILENO
]);
4049 if (c
->std_output
== EXEC_OUTPUT_FILE
)
4050 fprintf(f
, "%sStandardOutputFile: %s\n", prefix
, c
->stdio_file
[STDOUT_FILENO
]);
4051 if (c
->std_error
== EXEC_OUTPUT_FILE
)
4052 fprintf(f
, "%sStandardErrorFile: %s\n", prefix
, c
->stdio_file
[STDERR_FILENO
]);
4058 "%sTTYVHangup: %s\n"
4059 "%sTTYVTDisallocate: %s\n",
4060 prefix
, c
->tty_path
,
4061 prefix
, yes_no(c
->tty_reset
),
4062 prefix
, yes_no(c
->tty_vhangup
),
4063 prefix
, yes_no(c
->tty_vt_disallocate
));
4065 if (IN_SET(c
->std_output
,
4068 EXEC_OUTPUT_JOURNAL
,
4069 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4070 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4071 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
) ||
4072 IN_SET(c
->std_error
,
4075 EXEC_OUTPUT_JOURNAL
,
4076 EXEC_OUTPUT_SYSLOG_AND_CONSOLE
,
4077 EXEC_OUTPUT_KMSG_AND_CONSOLE
,
4078 EXEC_OUTPUT_JOURNAL_AND_CONSOLE
)) {
4080 _cleanup_free_
char *fac_str
= NULL
, *lvl_str
= NULL
;
4082 r
= log_facility_unshifted_to_string_alloc(c
->syslog_priority
>> 3, &fac_str
);
4084 fprintf(f
, "%sSyslogFacility: %s\n", prefix
, fac_str
);
4086 r
= log_level_to_string_alloc(LOG_PRI(c
->syslog_priority
), &lvl_str
);
4088 fprintf(f
, "%sSyslogLevel: %s\n", prefix
, lvl_str
);
4091 if (c
->log_level_max
>= 0) {
4092 _cleanup_free_
char *t
= NULL
;
4094 (void) log_level_to_string_alloc(c
->log_level_max
, &t
);
4096 fprintf(f
, "%sLogLevelMax: %s\n", prefix
, strna(t
));
4099 if (c
->n_log_extra_fields
> 0) {
4102 for (j
= 0; j
< c
->n_log_extra_fields
; j
++) {
4103 fprintf(f
, "%sLogExtraFields: ", prefix
);
4104 fwrite(c
->log_extra_fields
[j
].iov_base
,
4105 1, c
->log_extra_fields
[j
].iov_len
,
4111 if (c
->secure_bits
) {
4112 _cleanup_free_
char *str
= NULL
;
4114 r
= secure_bits_to_string_alloc(c
->secure_bits
, &str
);
4116 fprintf(f
, "%sSecure Bits: %s\n", prefix
, str
);
4119 if (c
->capability_bounding_set
!= CAP_ALL
) {
4120 _cleanup_free_
char *str
= NULL
;
4122 r
= capability_set_to_string_alloc(c
->capability_bounding_set
, &str
);
4124 fprintf(f
, "%sCapabilityBoundingSet: %s\n", prefix
, str
);
4127 if (c
->capability_ambient_set
!= 0) {
4128 _cleanup_free_
char *str
= NULL
;
4130 r
= capability_set_to_string_alloc(c
->capability_ambient_set
, &str
);
4132 fprintf(f
, "%sAmbientCapabilities: %s\n", prefix
, str
);
4136 fprintf(f
, "%sUser: %s\n", prefix
, c
->user
);
4138 fprintf(f
, "%sGroup: %s\n", prefix
, c
->group
);
4140 fprintf(f
, "%sDynamicUser: %s\n", prefix
, yes_no(c
->dynamic_user
));
4142 if (!strv_isempty(c
->supplementary_groups
)) {
4143 fprintf(f
, "%sSupplementaryGroups:", prefix
);
4144 strv_fprintf(f
, c
->supplementary_groups
);
4149 fprintf(f
, "%sPAMName: %s\n", prefix
, c
->pam_name
);
4151 if (!strv_isempty(c
->read_write_paths
)) {
4152 fprintf(f
, "%sReadWritePaths:", prefix
);
4153 strv_fprintf(f
, c
->read_write_paths
);
4157 if (!strv_isempty(c
->read_only_paths
)) {
4158 fprintf(f
, "%sReadOnlyPaths:", prefix
);
4159 strv_fprintf(f
, c
->read_only_paths
);
4163 if (!strv_isempty(c
->inaccessible_paths
)) {
4164 fprintf(f
, "%sInaccessiblePaths:", prefix
);
4165 strv_fprintf(f
, c
->inaccessible_paths
);
4169 if (c
->n_bind_mounts
> 0)
4170 for (i
= 0; i
< c
->n_bind_mounts
; i
++)
4171 fprintf(f
, "%s%s: %s%s:%s:%s\n", prefix
,
4172 c
->bind_mounts
[i
].read_only
? "BindReadOnlyPaths" : "BindPaths",
4173 c
->bind_mounts
[i
].ignore_enoent
? "-": "",
4174 c
->bind_mounts
[i
].source
,
4175 c
->bind_mounts
[i
].destination
,
4176 c
->bind_mounts
[i
].recursive
? "rbind" : "norbind");
4178 if (c
->n_temporary_filesystems
> 0)
4179 for (i
= 0; i
< c
->n_temporary_filesystems
; i
++) {
4180 TemporaryFileSystem
*t
= c
->temporary_filesystems
+ i
;
4182 fprintf(f
, "%sTemporaryFileSystem: %s%s%s\n", prefix
,
4184 isempty(t
->options
) ? "" : ":",
4185 strempty(t
->options
));
4190 "%sUtmpIdentifier: %s\n",
4191 prefix
, c
->utmp_id
);
4193 if (c
->selinux_context
)
4195 "%sSELinuxContext: %s%s\n",
4196 prefix
, c
->selinux_context_ignore
? "-" : "", c
->selinux_context
);
4198 if (c
->apparmor_profile
)
4200 "%sAppArmorProfile: %s%s\n",
4201 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4203 if (c
->smack_process_label
)
4205 "%sSmackProcessLabel: %s%s\n",
4206 prefix
, c
->smack_process_label_ignore
? "-" : "", c
->smack_process_label
);
4208 if (c
->personality
!= PERSONALITY_INVALID
)
4210 "%sPersonality: %s\n",
4211 prefix
, strna(personality_to_string(c
->personality
)));
4214 "%sLockPersonality: %s\n",
4215 prefix
, yes_no(c
->lock_personality
));
4217 if (c
->syscall_filter
) {
4225 "%sSystemCallFilter: ",
4228 if (!c
->syscall_whitelist
)
4232 HASHMAP_FOREACH_KEY(val
, id
, c
->syscall_filter
, j
) {
4233 _cleanup_free_
char *name
= NULL
;
4234 const char *errno_name
= NULL
;
4235 int num
= PTR_TO_INT(val
);
4242 name
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, PTR_TO_INT(id
) - 1);
4243 fputs(strna(name
), f
);
4246 errno_name
= errno_to_name(num
);
4248 fprintf(f
, ":%s", errno_name
);
4250 fprintf(f
, ":%d", num
);
4258 if (c
->syscall_archs
) {
4265 "%sSystemCallArchitectures:",
4269 SET_FOREACH(id
, c
->syscall_archs
, j
)
4270 fprintf(f
, " %s", strna(seccomp_arch_to_string(PTR_TO_UINT32(id
) - 1)));
4275 if (exec_context_restrict_namespaces_set(c
)) {
4276 _cleanup_free_
char *s
= NULL
;
4278 r
= namespace_flag_to_string_many(c
->restrict_namespaces
, &s
);
4280 fprintf(f
, "%sRestrictNamespaces: %s\n",
4284 if (c
->syscall_errno
> 0) {
4285 const char *errno_name
;
4287 fprintf(f
, "%sSystemCallErrorNumber: ", prefix
);
4289 errno_name
= errno_to_name(c
->syscall_errno
);
4291 fprintf(f
, "%s\n", errno_name
);
4293 fprintf(f
, "%d\n", c
->syscall_errno
);
4296 if (c
->apparmor_profile
)
4298 "%sAppArmorProfile: %s%s\n",
4299 prefix
, c
->apparmor_profile_ignore
? "-" : "", c
->apparmor_profile
);
4302 bool exec_context_maintains_privileges(const ExecContext
*c
) {
4305 /* Returns true if the process forked off would run under
4306 * an unchanged UID or as root. */
4311 if (streq(c
->user
, "root") || streq(c
->user
, "0"))
4317 int exec_context_get_effective_ioprio(const ExecContext
*c
) {
4325 p
= ioprio_get(IOPRIO_WHO_PROCESS
, 0);
4327 return IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE
, 4);
4332 void exec_context_free_log_extra_fields(ExecContext
*c
) {
4337 for (l
= 0; l
< c
->n_log_extra_fields
; l
++)
4338 free(c
->log_extra_fields
[l
].iov_base
);
4339 c
->log_extra_fields
= mfree(c
->log_extra_fields
);
4340 c
->n_log_extra_fields
= 0;
4343 void exec_status_start(ExecStatus
*s
, pid_t pid
) {
4348 dual_timestamp_get(&s
->start_timestamp
);
4351 void exec_status_exit(ExecStatus
*s
, const ExecContext
*context
, pid_t pid
, int code
, int status
) {
4354 if (s
->pid
&& s
->pid
!= pid
)
4358 dual_timestamp_get(&s
->exit_timestamp
);
4364 if (context
->utmp_id
)
4365 utmp_put_dead_process(context
->utmp_id
, pid
, code
, status
);
4367 exec_context_tty_reset(context
, NULL
);
4371 void exec_status_dump(const ExecStatus
*s
, FILE *f
, const char *prefix
) {
4372 char buf
[FORMAT_TIMESTAMP_MAX
];
4380 prefix
= strempty(prefix
);
4383 "%sPID: "PID_FMT
"\n",
4386 if (dual_timestamp_is_set(&s
->start_timestamp
))
4388 "%sStart Timestamp: %s\n",
4389 prefix
, format_timestamp(buf
, sizeof(buf
), s
->start_timestamp
.realtime
));
4391 if (dual_timestamp_is_set(&s
->exit_timestamp
))
4393 "%sExit Timestamp: %s\n"
4395 "%sExit Status: %i\n",
4396 prefix
, format_timestamp(buf
, sizeof(buf
), s
->exit_timestamp
.realtime
),
4397 prefix
, sigchld_code_to_string(s
->code
),
4401 static char *exec_command_line(char **argv
) {
4409 STRV_FOREACH(a
, argv
)
4417 STRV_FOREACH(a
, argv
) {
4424 if (strpbrk(*a
, WHITESPACE
)) {
4435 /* FIXME: this doesn't really handle arguments that have
4436 * spaces and ticks in them */
4441 static void exec_command_dump(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4442 _cleanup_free_
char *cmd
= NULL
;
4443 const char *prefix2
;
4448 prefix
= strempty(prefix
);
4449 prefix2
= strjoina(prefix
, "\t");
4451 cmd
= exec_command_line(c
->argv
);
4453 "%sCommand Line: %s\n",
4454 prefix
, cmd
? cmd
: strerror(ENOMEM
));
4456 exec_status_dump(&c
->exec_status
, f
, prefix2
);
4459 void exec_command_dump_list(ExecCommand
*c
, FILE *f
, const char *prefix
) {
4462 prefix
= strempty(prefix
);
4464 LIST_FOREACH(command
, c
, c
)
4465 exec_command_dump(c
, f
, prefix
);
4468 void exec_command_append_list(ExecCommand
**l
, ExecCommand
*e
) {
4475 /* It's kind of important, that we keep the order here */
4476 LIST_FIND_TAIL(command
, *l
, end
);
4477 LIST_INSERT_AFTER(command
, *l
, end
, e
);
4482 int exec_command_set(ExecCommand
*c
, const char *path
, ...) {
4490 l
= strv_new_ap(path
, ap
);
4511 int exec_command_append(ExecCommand
*c
, const char *path
, ...) {
4512 _cleanup_strv_free_
char **l
= NULL
;
4520 l
= strv_new_ap(path
, ap
);
4526 r
= strv_extend_strv(&c
->argv
, l
, false);
4533 static void *remove_tmpdir_thread(void *p
) {
4534 _cleanup_free_
char *path
= p
;
4536 (void) rm_rf(path
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
4540 static ExecRuntime
* exec_runtime_free(ExecRuntime
*rt
, bool destroy
) {
4547 (void) hashmap_remove(rt
->manager
->exec_runtime_by_id
, rt
->id
);
4549 /* When destroy is true, then rm_rf tmp_dir and var_tmp_dir. */
4550 if (destroy
&& rt
->tmp_dir
) {
4551 log_debug("Spawning thread to nuke %s", rt
->tmp_dir
);
4553 r
= asynchronous_job(remove_tmpdir_thread
, rt
->tmp_dir
);
4555 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->tmp_dir
);
4562 if (destroy
&& rt
->var_tmp_dir
) {
4563 log_debug("Spawning thread to nuke %s", rt
->var_tmp_dir
);
4565 r
= asynchronous_job(remove_tmpdir_thread
, rt
->var_tmp_dir
);
4567 log_warning_errno(r
, "Failed to nuke %s: %m", rt
->var_tmp_dir
);
4568 free(rt
->var_tmp_dir
);
4571 rt
->var_tmp_dir
= NULL
;
4574 rt
->id
= mfree(rt
->id
);
4575 rt
->tmp_dir
= mfree(rt
->tmp_dir
);
4576 rt
->var_tmp_dir
= mfree(rt
->var_tmp_dir
);
4577 safe_close_pair(rt
->netns_storage_socket
);
4581 static void exec_runtime_freep(ExecRuntime
**rt
) {
4583 (void) exec_runtime_free(*rt
, false);
4586 static int exec_runtime_allocate(ExecRuntime
**rt
) {
4589 *rt
= new0(ExecRuntime
, 1);
4593 (*rt
)->netns_storage_socket
[0] = (*rt
)->netns_storage_socket
[1] = -1;
4597 static int exec_runtime_add(
4600 const char *tmp_dir
,
4601 const char *var_tmp_dir
,
4602 const int netns_storage_socket
[2],
4603 ExecRuntime
**ret
) {
4605 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt
= NULL
;
4611 r
= hashmap_ensure_allocated(&m
->exec_runtime_by_id
, &string_hash_ops
);
4615 r
= exec_runtime_allocate(&rt
);
4619 rt
->id
= strdup(id
);
4624 rt
->tmp_dir
= strdup(tmp_dir
);
4628 /* When tmp_dir is set, then we require var_tmp_dir is also set. */
4629 assert(var_tmp_dir
);
4630 rt
->var_tmp_dir
= strdup(var_tmp_dir
);
4631 if (!rt
->var_tmp_dir
)
4635 if (netns_storage_socket
) {
4636 rt
->netns_storage_socket
[0] = netns_storage_socket
[0];
4637 rt
->netns_storage_socket
[1] = netns_storage_socket
[1];
4640 r
= hashmap_put(m
->exec_runtime_by_id
, rt
->id
, rt
);
4649 /* do not remove created ExecRuntime object when the operation succeeds. */
4654 static int exec_runtime_make(Manager
*m
, const ExecContext
*c
, const char *id
, ExecRuntime
**ret
) {
4655 _cleanup_free_
char *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4656 _cleanup_close_pair_
int netns_storage_socket
[2] = {-1, -1};
4663 /* It is not necessary to create ExecRuntime object. */
4664 if (!c
->private_network
&& !c
->private_tmp
)
4667 if (c
->private_tmp
) {
4668 r
= setup_tmp_dirs(id
, &tmp_dir
, &var_tmp_dir
);
4673 if (c
->private_network
) {
4674 if (socketpair(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
, 0, netns_storage_socket
) < 0)
4678 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, netns_storage_socket
, ret
);
4683 netns_storage_socket
[0] = -1;
4684 netns_storage_socket
[1] = -1;
4688 int exec_runtime_acquire(Manager
*m
, const ExecContext
*c
, const char *id
, bool create
, ExecRuntime
**ret
) {
4696 rt
= hashmap_get(m
->exec_runtime_by_id
, id
);
4698 /* We already have a ExecRuntime object, let's increase the ref count and reuse it */
4704 /* If not found, then create a new object. */
4705 r
= exec_runtime_make(m
, c
, id
, &rt
);
4707 /* When r == 0, it is not necessary to create ExecRuntime object. */
4711 /* increment reference counter. */
4717 ExecRuntime
*exec_runtime_unref(ExecRuntime
*rt
, bool destroy
) {
4721 assert(rt
->n_ref
> 0);
4727 return exec_runtime_free(rt
, destroy
);
4730 int exec_runtime_serialize(const Manager
*m
, FILE *f
, FDSet
*fds
) {
4738 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
4739 fprintf(f
, "exec-runtime=%s", rt
->id
);
4742 fprintf(f
, " tmp-dir=%s", rt
->tmp_dir
);
4744 if (rt
->var_tmp_dir
)
4745 fprintf(f
, " var-tmp-dir=%s", rt
->var_tmp_dir
);
4747 if (rt
->netns_storage_socket
[0] >= 0) {
4750 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[0]);
4754 fprintf(f
, " netns-socket-0=%i", copy
);
4757 if (rt
->netns_storage_socket
[1] >= 0) {
4760 copy
= fdset_put_dup(fds
, rt
->netns_storage_socket
[1]);
4764 fprintf(f
, " netns-socket-1=%i", copy
);
4773 int exec_runtime_deserialize_compat(Unit
*u
, const char *key
, const char *value
, FDSet
*fds
) {
4774 _cleanup_(exec_runtime_freep
) ExecRuntime
*rt_create
= NULL
;
4778 /* This is for the migration from old (v237 or earlier) deserialization text.
4779 * Due to the bug #7790, this may not work with the units that use JoinsNamespaceOf=.
4780 * Even if the ExecRuntime object originally created by the other unit, we cannot judge
4781 * so or not from the serialized text, then we always creates a new object owned by this. */
4787 /* Manager manages ExecRuntime objects by the unit id.
4788 * So, we omit the serialized text when the unit does not have id (yet?)... */
4789 if (isempty(u
->id
)) {
4790 log_unit_debug(u
, "Invocation ID not found. Dropping runtime parameter.");
4794 r
= hashmap_ensure_allocated(&u
->manager
->exec_runtime_by_id
, &string_hash_ops
);
4796 log_unit_debug_errno(u
, r
, "Failed to allocate storage for runtime parameter: %m");
4800 rt
= hashmap_get(u
->manager
->exec_runtime_by_id
, u
->id
);
4802 r
= exec_runtime_allocate(&rt_create
);
4806 rt_create
->id
= strdup(u
->id
);
4813 if (streq(key
, "tmp-dir")) {
4816 copy
= strdup(value
);
4820 free_and_replace(rt
->tmp_dir
, copy
);
4822 } else if (streq(key
, "var-tmp-dir")) {
4825 copy
= strdup(value
);
4829 free_and_replace(rt
->var_tmp_dir
, copy
);
4831 } else if (streq(key
, "netns-socket-0")) {
4834 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
4835 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4839 safe_close(rt
->netns_storage_socket
[0]);
4840 rt
->netns_storage_socket
[0] = fdset_remove(fds
, fd
);
4842 } else if (streq(key
, "netns-socket-1")) {
4845 if (safe_atoi(value
, &fd
) < 0 || !fdset_contains(fds
, fd
)) {
4846 log_unit_debug(u
, "Failed to parse netns socket value: %s", value
);
4850 safe_close(rt
->netns_storage_socket
[1]);
4851 rt
->netns_storage_socket
[1] = fdset_remove(fds
, fd
);
4856 /* If the object is newly created, then put it to the hashmap which manages ExecRuntime objects. */
4858 r
= hashmap_put(u
->manager
->exec_runtime_by_id
, rt_create
->id
, rt_create
);
4860 log_unit_debug_errno(u
, r
, "Failed to put runtime paramter to manager's storage: %m");
4864 rt_create
->manager
= u
->manager
;
4873 void exec_runtime_deserialize_one(Manager
*m
, const char *value
, FDSet
*fds
) {
4874 char *id
= NULL
, *tmp_dir
= NULL
, *var_tmp_dir
= NULL
;
4875 int r
, fd0
= -1, fd1
= -1;
4876 const char *p
, *v
= value
;
4883 n
= strcspn(v
, " ");
4884 id
= strndupa(v
, n
);
4889 v
= startswith(p
, "tmp-dir=");
4891 n
= strcspn(v
, " ");
4892 tmp_dir
= strndupa(v
, n
);
4898 v
= startswith(p
, "var-tmp-dir=");
4900 n
= strcspn(v
, " ");
4901 var_tmp_dir
= strndupa(v
, n
);
4907 v
= startswith(p
, "netns-socket-0=");
4911 n
= strcspn(v
, " ");
4912 buf
= strndupa(v
, n
);
4913 if (safe_atoi(buf
, &fd0
) < 0 || !fdset_contains(fds
, fd0
)) {
4914 log_debug("Unable to process exec-runtime netns fd specification.");
4917 fd0
= fdset_remove(fds
, fd0
);
4923 v
= startswith(p
, "netns-socket-1=");
4927 n
= strcspn(v
, " ");
4928 buf
= strndupa(v
, n
);
4929 if (safe_atoi(buf
, &fd1
) < 0 || !fdset_contains(fds
, fd1
)) {
4930 log_debug("Unable to process exec-runtime netns fd specification.");
4933 fd1
= fdset_remove(fds
, fd1
);
4938 r
= exec_runtime_add(m
, id
, tmp_dir
, var_tmp_dir
, (int[]) { fd0
, fd1
}, NULL
);
4940 log_debug_errno(r
, "Failed to add exec-runtime: %m");
4945 void exec_runtime_vacuum(Manager
*m
) {
4951 /* Free unreferenced ExecRuntime objects. This is used after manager deserialization process. */
4953 HASHMAP_FOREACH(rt
, m
->exec_runtime_by_id
, i
) {
4957 (void) exec_runtime_free(rt
, false);
4961 static const char* const exec_input_table
[_EXEC_INPUT_MAX
] = {
4962 [EXEC_INPUT_NULL
] = "null",
4963 [EXEC_INPUT_TTY
] = "tty",
4964 [EXEC_INPUT_TTY_FORCE
] = "tty-force",
4965 [EXEC_INPUT_TTY_FAIL
] = "tty-fail",
4966 [EXEC_INPUT_SOCKET
] = "socket",
4967 [EXEC_INPUT_NAMED_FD
] = "fd",
4968 [EXEC_INPUT_DATA
] = "data",
4969 [EXEC_INPUT_FILE
] = "file",
4972 DEFINE_STRING_TABLE_LOOKUP(exec_input
, ExecInput
);
4974 static const char* const exec_output_table
[_EXEC_OUTPUT_MAX
] = {
4975 [EXEC_OUTPUT_INHERIT
] = "inherit",
4976 [EXEC_OUTPUT_NULL
] = "null",
4977 [EXEC_OUTPUT_TTY
] = "tty",
4978 [EXEC_OUTPUT_SYSLOG
] = "syslog",
4979 [EXEC_OUTPUT_SYSLOG_AND_CONSOLE
] = "syslog+console",
4980 [EXEC_OUTPUT_KMSG
] = "kmsg",
4981 [EXEC_OUTPUT_KMSG_AND_CONSOLE
] = "kmsg+console",
4982 [EXEC_OUTPUT_JOURNAL
] = "journal",
4983 [EXEC_OUTPUT_JOURNAL_AND_CONSOLE
] = "journal+console",
4984 [EXEC_OUTPUT_SOCKET
] = "socket",
4985 [EXEC_OUTPUT_NAMED_FD
] = "fd",
4986 [EXEC_OUTPUT_FILE
] = "file",
4989 DEFINE_STRING_TABLE_LOOKUP(exec_output
, ExecOutput
);
4991 static const char* const exec_utmp_mode_table
[_EXEC_UTMP_MODE_MAX
] = {
4992 [EXEC_UTMP_INIT
] = "init",
4993 [EXEC_UTMP_LOGIN
] = "login",
4994 [EXEC_UTMP_USER
] = "user",
4997 DEFINE_STRING_TABLE_LOOKUP(exec_utmp_mode
, ExecUtmpMode
);
4999 static const char* const exec_preserve_mode_table
[_EXEC_PRESERVE_MODE_MAX
] = {
5000 [EXEC_PRESERVE_NO
] = "no",
5001 [EXEC_PRESERVE_YES
] = "yes",
5002 [EXEC_PRESERVE_RESTART
] = "restart",
5005 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(exec_preserve_mode
, ExecPreserveMode
, EXEC_PRESERVE_YES
);
5007 static const char* const exec_directory_type_table
[_EXEC_DIRECTORY_TYPE_MAX
] = {
5008 [EXEC_DIRECTORY_RUNTIME
] = "RuntimeDirectory",
5009 [EXEC_DIRECTORY_STATE
] = "StateDirectory",
5010 [EXEC_DIRECTORY_CACHE
] = "CacheDirectory",
5011 [EXEC_DIRECTORY_LOGS
] = "LogsDirectory",
5012 [EXEC_DIRECTORY_CONFIGURATION
] = "ConfigurationDirectory",
5015 DEFINE_STRING_TABLE_LOOKUP(exec_directory_type
, ExecDirectoryType
);
5017 static const char* const exec_keyring_mode_table
[_EXEC_KEYRING_MODE_MAX
] = {
5018 [EXEC_KEYRING_INHERIT
] = "inherit",
5019 [EXEC_KEYRING_PRIVATE
] = "private",
5020 [EXEC_KEYRING_SHARED
] = "shared",
5023 DEFINE_STRING_TABLE_LOOKUP(exec_keyring_mode
, ExecKeyringMode
);